Uncategorized

Okay, welcome, everybody This is, um, like an interesting time, like, to be here Like, um, actually, we got, like, a great, like, day here in Boston, which actually is not something that is, um, it has to be taken lightly, yeah? Um, first of all, let me thank you for being here. I think it’s, like, our pleasure to share this time with you guys We are going to present this day a very interesting work about how to mix two things that apparently don’t do much, but together, they do something really interesting, right? So, before we really start, let’s do a little bit of housekeeping. So, the day’s organized in three sessions The first session is going to be a session of, like, keynotes, so we have people like Sandy Pentland, like Alex, Stepahine, people that are at the forefront of their fields, but also, we are going to have a second session in the afternoon in which we are going to have presenters of papers, authors, that submitted their work to this symposium and that are basically trying to tackle a problem within this space, right? And then at the end, in the afternoon, we are going to have a third session in which we’re going to have an actual demo, people from DCI is going to bring here a robot, a vending machine actually that runs like a blockchain, and they’re going to teach us, like, how it operates, and then we’re going to finish the symposium with a panel discussion with Neh a Narula, the head of DCI So, first of all, let’s thank the committee members who reviewed these papers, so we have people, um, I’d just like to give you data about how this process went. We received, like, 20 initial submissions, which is a lot of submissions for this, like, new field We ended up accepting eight papers, so we had a 40 percent acceptance rate, and what you can see here in the graphics basically, the page views of our, like, submission site and the website that you all have seen, so we got some attention, right, for being such a novel topic. So, I am happy to announce that these submissions that got accepted are going to be published in the Ledger Journal, which is like a new, um, open access journal that’s just, like, open, it’s just got, like, two or three issues, and we expect to have a special issue for this symposium early in 2019. So, with this, let me start my introduction and my welcome, right? So, you all know that we live in a world which is, um, pretty diverse, right? And there’s, I don’t know if people think, like, in many different ways about the world of robotics, AI, etc., right? I’m sure that you all have spoken with, like, somebody that says that robots are going to come, robots are going to take over, like, they are the start of, like, a course for humanity, and you also talk to people that think that this is not like that, right? So, robots are going to help us, robots are not here to substitute us, they’re here to extend us, right? This is a difficult topic, right? But let me give you a little bit of context. So, during my master’s and my Ph D degree in Japan, I work for one of these two guys here, right? So, one was, like, my supervisor and professor, the other was, like, his clone, yeah? His proxy So, basically, Professor tried this very heterogeneous and fancy way, like, to solve a problem, think about how he could be in two places at the same time. He decided, like, to create a clone and send the clone wherever he didn’t want to be, and then, like, basically, look through this, like, robot and give lectures, presentations, etc., through him So, I, maybe, like, you can just see it in this video, this is a video in which the two of them are in the same room, so it doesn’t fulfill this, like, teleoperation purpose, but more or less, you can understand what’s going on here So, he has, like, some kind of, um, motion recognition, and this motion recognition, like from the muscles, the face, etc., the voice, is, like, being transferred to the other, right? You would have thought that the future of robotics is this, the fact that you will have a very expensive robot next to you, and then you will be able to send it wherever you don’t want to be. After working with him, like, eight years, I realized that, actually, this is

not the vision that I had, you know, for the future of robotics. My vision was something a little bit different. Instead of having one very complex robot next to you, I thoughtthat the future of robotics was having many different robots around you, robots that could do simple, robots that could do maybe one thing, but they could do it right, but through collaborating with other robots, could do very, very complex tasks. So, basically, what you are seeing here is a little swarm of robots doing something called foraging, in which the robots have to collaborate in order to get these, like, 3D printed tokens that you can see This is me stealing the food of the robots. Then, like, trying to organize themselves in order to put these three tokens in In exchange of these tokens, it would give some kind of, like, battery recharge to the robots So, we were trying to create a loop of control in which the robots were able to self-organize and conduct something for really, really long periods of time. So, instead of having, like, three printed tokens, if you see, for example, trash or minerals, etc , this model starts to get a little bit more sense, right? So, when I finished my Ph D, I realized that there were other models, there were other, like, um, groups of people collaborating and corporating to do something beyond the individual, right? That’s why I decided to come here and join the human dynamics group, because in the group, we have a lot of research about how these human societies cooperate ask and collaborate on Earth to do these tasks. So, I joined the group about one year ago and started to do research about this, but then I realized one thing, if you are going to have many, many robots around us, this could be perfectly, so, your smart phone could be like a robot, your smart home could be a robot, your smart car, right? There’s a problem here, and the problem is that through having a lot of devices around us and making them talking to each other, the complexity of the network, like, around us increases Up to a certainly level, that could make our cognit cognitive ability not enough to understand what’s going around us. So, for example, in this graph, what we can see is that as the number of devices inCroesuses increases, the complexity of that network also increases. If you have, like, the same type of device, like, for example, yeah, it could increase linearly, but if you have a heterogeneous network of your smart robot with your smart car with your smart agenda with your smart phone, it could increase exponentially Up to a certainly level, it could actually surpass our cognitive ability to understand what’s going on there. So, I realized through connecting research in the group that it would be fantastic, yeah, like, to have, like, a certain tool that could actually make us have, like, really complex networks of devices, like, around decentralized networks, but at the same time, having some kind of, like, tool, transparent tool, to actually not surpass our cognitive abilities and keep us in the pace of understanding what’s going on around us One of the key things for that, one of the key things in order to, like, reach this, like, green line is the fact of having trust, but the problem is that trust is something that we more or less can understand, I am pretty sure that you are who you are, right, and, like, this chair is this chair and I’m a speaker, but in this world, trust is a very tricky thing, especially because the senate of the Internet, the tool that we all use, was not designed initially in order to convey trust It was designed in order to, like, move efficiently data from point A to point B, but not to generate trust, and that’s one of the origins of all the problems that we have these days. So, I am completely sure that I don’t need to explain to you what the blockchain is Actually, I should say the blockchain, yeah, and I say the blockchain instead of blockchain, because as you all know, blockchain is kind of, like, a bash word, right? It means everything and nothing at the same time, no? So, what I’m going to explain to you in these slides is not actually how the blockchain will solve all our problems, but how certain tools within this concept ofblack could be helpful with the systems we have today, right? And one of these key things, one of these key ideas of this blockchain is the fact that, like now, we can have, like, a source of trust from centralized perspective to a decentralized perspective, right? So, in other words, if, at some point

in time, some value, for example, like 10ureoes, $10, 10 bit coins, in the world, up to now, you needed to have some centralized point of trust, a government, a bank, in order to vouch for that value, right, in order to evaluate all these transactions The blockchain, the only thing that it did is passing from this, that centralized model to this decentralized model. Now, we don’t need, like, a middle man to vouch, like, for that value. If we have a network of people that maybe don’t trust each other, maybe don’t live in the same place, but they agree on a certain point in time, that some transaction passed, or some transaction happened, or some value moved This is the main revolution of this blockchain. So, these transactions, okay, they are moving, like, all over the network, are incapsulated in blocks, like many of you will know, and at some point in time, somebody signs these blocks crypt graphically, right? So, it can account for the fact these transactions happened, and redoing this transaction or altering it is extremely costly, and that’s one of the key ideas of this thing. In a way, what we have, like, with this, um, with blockchain is that we have, like, a history, we created a history book of the transactions that happened within a group of peers, right, a group of agents. What I’m proposing and what I’m going to show you in the next slide is what happens when you have this history book, not with a group of people, but with a group of robots, what can you achieve, and this all started in 2016, right? Because, um, I realized that in my world, in a world of robotics and especially in a world of distributed robotics, there were certain problems that needed to be addressed in order to take this world from academic peer research to industry, and then I realized that this blockchain could be a good tool in order, like, to bridge the gap. So, I wrote this article, yeah, that maybe some of you, like, have read, and this, a little bit started, like, this field that we are going to discuss today In that article, what I commented is that in order to pass from pure academic research in the world of robotics and distributed robotics and going, like, to adoption, mainstream adoption, like, industry, etc , we need to solve these four problems. We need to solve the problem of security with these systems, how do we create, like, tools in order to, like, account for these systems, how do we get new consensus problems, how do we make robots not only talk to each other, but agree on certain things, have, like, different behaviors, so the robots or groups of robots could do two things at the same time, and, of course, having new business models. So, one of the things that this, like, article said in 2016 is the fact that even though there’s people talking, like, about how robots are going, like, to hit our lives and are going to be part of our daily activities, if we don’t care about how do we secure robots, things are going to be, could go south, right? So, maybe many of you have heard about this, malware or ransomware, where you opened up a message that said if you don’t pay me $300 in bit coin, your data is going to be lost. Actually, it was not fun, but that was the start of something more profound Let’s imagine that that ransomware not only affects your computer, but affects your smart phone, affects, for example, your self-driving car. Could you imagine in the future, when we have self-driving cars, that at some point in time, you are on the highway, and some screen pops up and says if you don’t transfer me $3,000 in bit coin, I’m going to break? Yeah, things get a little weirder, but this is not the future Actually, this is present, yeah? So, maybe you know, like, these robots are implemented in certain hospitals here in the United States, and this robot was operating, like, in groups in order to deliver, um, medical records, like, medicines to certain patients, help nurses, etc., right? So, they designed, like, this group of robots in order to exchange information with one another, etc , right? You have many members, so if one robot breaks, the others could take over, right? But what they didn’t realize is what happens not when one robot breaks, but when one robot

deceives the other robots, when you had one robot that is, like, programmed in order to poison the communications with the other robots Yes, I visited room 609, don’t worry, you don’t have to go there. Well, it’s proven through this example that if you hack one robot in a swarm, eventually, the whole swarm will collapse. So, when you hear, like, about Amazon delivering packages, like, through drones or swarms of, like, self-driving cars in cities, think about that So, okay, so what we do here at MIT, we understood that there was a problem, and then we developed technology, like, in order, like, to solve it So, what you can see here is basically one of the simplest problems you can, um, see in the world of distributed robotics, which is, like, the consensus, right? So, what you’re seeing here is, like, a swarm of robots, completely centralized, which are, like, wondering around this, like, checkerboard, and they are super simple, the only thing that they can do is just, like, sense if they are in a white tile or black tile, but they have some memory, so they can just have a percentage of how many white tiles over black tiles. So, once, like, they bump into another robot, they are able to exchange opinions and opinions, but at some point in time, there’s a consensus. The robots without centralized controller can get the opinion of, oh, we think that the majority controller in this checkerboard is white or black. So, what we did is, like, putting these robots, increasing the difficulty of the checkerboard, we saw how the robots behaved, and then we said, okay, so let’s put to test a blockchain approach in which they’re just, like, talking to each other plainly, they incapsulate their boats through blockchain transactions, so we can account them, we can see them, we see who bodes what, and we are going to compare to the state of the art algorithms Well, so, we run a test, and then we go to, like, these results. So, you see here, like, two graphs, these are different algorithms, in the X axis, you can see the difficulty of the environment, so, of course, 80/20, so 80 percent, like, black over 20 percent white is easier than, for example, 40/60, etc., so you see an increasing difficulty in the X axis, and you see the Y axis is the success probability, so the amount of time, so the probability that the robots achieved the correct solution at the correct time, right? Well, so, what you can see here is that if we compare the state of the art algorithm with a classical approach to the blockchain, there’s not much difference, right? So, um, we see that the success, like, rate is not different from one another, and, yeah, based on these results, we don’t have, like, much justification for putting blockchain there. Many of you know that putting out blockchain has its limitations, right? You know, like, these very simple robots, they have to keep all these bodes, they need to have a lot of communication overhead, right? They need to sign crypt graphically these transactions So, from these results, it doesn’t really make sense to put blockchain. Huh, but let’s imagine that we put, like, robots that are programmed to deceive the other robots, robots, for example, that simulate, like, a hardware problem or a sensor problem or they lost their communications, or actually, they tamper with the message that they want to broadcast, and then we do the same experiment So, the work flow of these experiments connect, like, an actuary mode, and then this mines, like, a contract, which basically is the Constitution of the robots, how they exchange bodes, how, like, um, what happens when we find out that a robot deceived the others, etc , and then this, like, contract serves as a kind of, like, publisher/subscriber mode, so robots can contract to these events, and then they get information back. So, um, at some point in time, this contract gets, like, mined into all the robots, and then the experiment starts. Well, here’s the same experiment. So, what we saw is that, for example, in the Y axis, in this case, we have the success probability, but in the X axis, we have the number of these robots, the number of robots that are programmed to deceive and break the consensus. So, what we can see is in the state of the art approach, as we increased more of these robots, the success probability drops dramatically. Even, for example, in a swarm of 25 members, the fact that we added three basintine robots makes this probability go to 0 1, and if we have four, it’s zero, impossible to achieve, but if we add

these, like, bodes into, like a, a blockchain, we are able to maintain the success probability pretty stable, and these results are explained with the following rationale. If I am a robot and then I go to you, and then I say the majority color is black, and then one minute afterwards, I went to you, and then I said, no, no, the majority color is white, and at some point in time, we all broke these transactions, at some point in time, we found an inconsistency, we found the fact that, actually, me, or us, having the same controller produced two different opinions within that timespan, for example, that is impossible for me, like, to get such a different opinion, right? So, finding these inconsistencies makes the group allow the robots, like, to say, well, maybe your reputation is not good enough, maybe there’s something wrong with you, maybe I’m not going to take your opinion into account Through finding inconsistencies that you will not find else where or you will not find in any other way, um, we are able to remove without the outliars and therefore keep the system running for longer periods of time. So, this is interesting, because this is the first, like, work that unites these two things, right? And then makes, like, a proof of concept that this technology is very interesting. So, this is the convergence time It’s actually the same story, right? But what this day is about is not only to show you, like, research that supports, like, the justification of mixing these two worlds, but the fact that we, what can we do when we can find a system or achieve something that’s accountable, right, that is trustable. So, if we can achieve such a system, maybe, in the future, we will be able, like, to see robots offering their services, right? For example, in this case, what you are seeing is what happens when a robot basically advertises itself on Facebook. What happens when robots have, like, advertise, like, their work, their services, etc This is very interesting, because it’s something we are currently working on, and this is a robot called Klouse, and basically, he’s collecting environmental data in the Antarctica. You might know that, like, researchers are now doing research on climate change and are passing through rough times, right? Yeah, and it’s basically because also, because many reasons, but one of the reasons is that it’s very costly to go and send a team to the Antarctica in order to, like, gather data, do experiments, etc., but for example, in the Antarctica, there’s already, like, robots like Klouse from NGOs that are being underused. So, what we propose is, like, a platform in which, like, these robots could advertise their services, their work, their output, when, for example, their owners are not, like, using them, right? So, in a sense, what we propose is, like, a system in which robots could federate themselves, yeah, based on their power, etc., etc , and you could basically hire them The interesting part is that once you are able, like, to get a robot that is where you want it, like, to be, it can provide the data, for example, that the robot, like, can provide and is usefulful, like, to you, you basically could negotiate that transaction into the blockchain and then make the robot, after its work, you know, deliver the data. Think about it, like, for example, in the case of Klouse, right, at some point in time, that researcher here, like at MIT or in a university, he what he really wants to get is the data, right? Afterwards, the only thing you want is the data, right? Data that the robot in the place, at the right time, at the right place, can provide, right? We just need an interface to make the transaction secure, but what I’m telling you is basically the projection of the world that we live today, if you think it properly, like, big players players today, like Uber, for example, owns no cars. Air B & B owns no real estate. Facebook generates no content whatsoever. And these players could be so big, because the paradigm is shifting, and it’s because, now, the access to information is more important than the ownership of that information So, what we think here is that that information that could be generated from our smart phones or could be generated from a robot, for example, is going to be in the market in the next 10 to 20 years, right? And the access to that information is going to be really important, but we need an interface in

order, like, to manage how we access that information, how we handle this information accessibility or utility, and also about the security of this information. So, that’s why, today, I would like to invite you to think about the synergy of three main concepts that are going to marriage together One, of course, is the world of robotics, of AI, automation, right? But the next one is going to be the world of digital trust. Now, we have techniques that, 10 years ago, we didn’t have them, and now, with the two worlds combined, we can create something that apparently doesn’t have to do much with each other, but together, does something interesting, but, of course, these two big, like, bubbles need to be placed somewhere, and this somewhere is going to be society, right? We need, like, to start, like, thinking about how we can make, like, these systems more accessible in order, like, to achieve a society that is, like, greener, more transparent, etc , and, of course, we are in a media lab, and one of the missions that we have here is to deploy systems, right? So, not only how to make, like, the synergy or explore the synergy of these bubbles, but also, like, to push it, like, to society, right, in order to do, like, a better society and have, like, higher impact So, with this, I think that my interaction is over, so I’m going to hand the presentation to Sandy Pentland. Thank you very much for being here (Applause ) >> SPEAKER: So, welcome Beautiful day. You guys are to be commended for sitting inside on a day like today So, I thought what I would do is I’d talk about, um, the research group that I run here at MIT and give sort of an overview of it, um, because it bears on the sort of questions, although most of us don’t do robots, we do people interacting and trust between them and contracts between, so maybe not so different Um, but I particularly wanted to, um, emphasize some of the things that aren’t widely known, which are ways of doing computation over, um, distributed and encrypted data, because it really changes the way, I think, you think about things, okay? So, um, this is me, I do lots of different things for the UN, for Google, people like that, and a lot of what I do is study people, privacy, security of data, because that’s sort of the big thing that’s happening out there, is data everywhere, and, um, we’ve developed a methodology of handling data that involves blockchain that we’ve been pushing as sort of a protostandard with a wide variety of companies, so some of these, Mastercard, Inter-America Development Bank, BBA, etc., but also many different countries, so Israel, France, Colombia, singal, China, are sponsors of ours to help develop ways to handle data, and some of these things, I think, are worth thinking about in the context of robots and blockchain. So, this is me, um, the President of the EU invited me to lecture all of the ministers of the single digital market in the fall, and I actually wore a suit and a tie, like almost never, right, and basically, what I told them was a couple of things Um, when I got there, what I heard from many of these ministers of the single digital market is they were concerned about GDPR, the privacy stuff, of course, but also about data localization, and the joke that everyone was saying was imagine you have an autonomous vehicle, and it drives from Germany into France, and suddenly, you had to go through explicit permissions for all of the information sources, and by that time, the car would run into a tree, and you’d die. So, what are we going to do about that? Which is fundamentally, um, a robot’s problem, um, and it’s sort of a near one to me, because I actually was involved in designing the autonomous, um, driving system for Nissan’s Altima, so I know a little bit about it The key thing, um, in the approach that we’ve developed, um, is that you don’t put data in one

spot. You want to be able to merge data dynamically and be able to control that merging, so share answers, not data. So, these are federated databases So, federated databases are unusual in computer science, because, um, in general, communication is much more expensive than computation or storage, but it turns out there’s lots of ways of doing that communication, computations across distributed things, that are efficient enough, and particularly with today’s technologies of fiber and 5G and things like that You know, it was shocking, the first time I learned what went on behind all of those ads that are on your browser, that they actually combine data sources from all over the planet in under 20 milliseconds to give you something. That just is a completely different way of thinking about, um, data systems, and that that really can be very distributed, if you’re careful about it So, the, um, place that blockchain comes in there is that when you build a federated data system, you have to be able to keep track of what you’re doing, it’s not the same as doing it in one spot. I should also mention, with a federated data system, so you have man adifferent actors, like different robots, um, you can’t compute everything typically, can’t do a complete drawing, is the database way of saying that, but typically, you don’t need to do that in an operational context. You might want to do that in research or development, but actually, for operation, it’s extremely unusual to have to do that sort of thing. So, what you’d like to be able to do is take data from all these different robots or all these different places and be able to ask questions of them, get answers, and you want to be able to log that on a blockchain, because all of the agents have to agree, or at least a majority of the agents have to agree that, you know, this was the correct credentials that I presented, this was versus what we put on the blockchain, is we actually put the algorithms for the various questions, so you know incontrovertibly what’s the code that’s being used, and then the answer is put up there too, perhaps with payments, and that can be encrypted or not encrypted, depending on the permission structure. What that does is that lets you be able to compute things across all these private databases, across all these robots, in a way that’s completely auditable and continuously auditable, and that’s one of the things Eddie was just talking about, is you have now the ability to watch a system, and because you have all the important parts of all of the important transactions, you can ask if it’s fair, if it’s achieving optimality, if it has mumalevolent people and so forth The introphist interesting thing is this general architecture was accepted about three weeks ago as the way the EU is going to run. All official statistics, all government statistics, all government databases will be moving to this type of a framework That’s a pretty amazing thing, right? Because who would have thought, but it’s clear what the motivation is. The motivation is if you have centralized nodes, they’re attackable, you lose everything. Nobody would think of doing that in military terms anymore, you don’t put your whole Army in one place, you distribute them, and that’s what we need to do here in this sort of modern world. So, what we’re moving from is we’re moving from a world where, you know, you have computation, and you think of it this way, you know, you get the password, you do something that can’t really be easily audited in there, or all too often is not audited, and then you get it output to something that looks like this, where I present my credentials, and my credentials should be distributed, so my identity should be a, it’s a primary key to the whole thing, you want, um, your identity to depend on attributions from many different domains, so that you can’t compromise and fake identities very easily. So, for instance, a majority of the other robots need to agree who I am before they’re going to accept my credentials. So, I present my credentials, along with my credentials, credentials come certain permissions, that gives me the ability to select an algorithm on the blockchain, that’s been put on the blockchain. You can’t put very much data on blockchains, right? That’s bad So, you have algorithms on the blockchain, you select ones, you

say run, you know, algorithm algorithm 23 on the local data of some robot or some data store, so it happens with the data in place. You don’t have multiple copies of the data, you’re not moving data, you’re doing it locally, it’s much safer in a variety of ways, and it solves a lot of the problems of who owns the data, who has rights to the data, because the data doesn’t move, right? You’re getting explicitly pre-agreed answers to questions, um, answers to explicitly peer-agreed questions, and that’s a much more controllable thing, and incidentally, in a practical system, you can actually get lawyers to agree to this, whereas if you say, well, I’m going to give you my data, and then you get to do things with it, a lawyer will never agree to that, because we don’t know what the risk involved is So, that’s the basic framework, and you can see that this is, like, the stuff that Eddie was talking about, which is why we work together. We’re doing this in a couple different things, places. For instance, in the state of Israel, we’re actually using this as a way of doing autonomous driving, so their innovation department has engaged us to setup a data-sharing platform for realtime data-sharing across many different companies, so that if you have a cool, new idea for an autonomous vehicle, you can now get realtime data from mobile I, from the bus companies, from the city city, from aerial things, all at the same time, without having to spend several million dollars on lawyers and all these sort of bad things, and it depends, of course, on these sorts of permission structures and this concept of not actually sharing the data, but sharing answers that are answers in a controlled sort of method It also gives you this ability to be able to monitor what’s happening, and that was this illustration here. For instance, in the countries of Colombia and Synegal, we’re setting up national data systems that involve people like Uber, the banks, and the government to be able to produce data that is census-like data, so, you know, where do people in this neighborhood move, you know, commute to, where do people in this neighborhood shop, how much do they spend, or what’s their capital, how rich are they, as a way of sort of monitoring the development then in the country and effective laws, and the fact that this is a distributed structure where the data doesn’t move and that it’s auditable lets you have a steering committee that watches it and can continuously monitor issues like fairness and bias, other sorts of things that might be relevant. Becomes not a serious problem to answer. You don’t care what the algorithms are so much as you care what the answers are, so being able to log all this stuff gives you the ability to sort of be able to control the ist system in a very different way. So, those two things. This is the third one Um, so, in the Snowden papers, you will notice, if you read them with a particular eye, that the moment you have data that is not encrypted, somebody steals it. So, what that means is that whatever system you build, that system that, the data that comes off of a censor sensor has to be immediately encrypted and never, never decrypted, which poses a problem, because you really want to do computation on it, right? Well, how do you do that? Well, it turns out that there’s a bunch of ways to do that without decrypting the data, and you can play with those words a little bit, and I’ll show you two ways of doing it Um, the classic way is, um, secure multi-party computing So, you may have heard of, um, I’m blocking on the name of it, um, the hot new way of doing this, but this is the classic way of doing this. Actually, what you do is you take these distributed databases, and you encrypt them in a very particular way that allows computation to be done on the encrypted data, so by sort of bit twiddling manipulations of the encrypted data, you can carry out certain, um, operations, and maybe you’re interested, I’ll give you the 60-second, you can do this at dinner version, if people want to hear that. It’s sort of fun,

right? This is like a card trick. See? Nothing up my sleeve, right? So, imagine that we want to do your salary, your salary, and my salary, and I want to know what the average of those are, right? But you don’t want to share your salary, I don’t want to share mine, how do we do this? And this is analogous to problems like, you know, are women paid as much as men, so you would do this across different databases. Nobody wants to reveal their data, because not only does it belong to the employee, but you could get sued, if it turns out that you’re not in good shape, right? So, you want to see averages Hospitals, how well does a particular procedure work, hospitals don’t want to reveal that data, it’s patient data, but more important to them is if it turns out they’re not very good at this operation, they’re going to lose money So, let’s do the salary So, you take your salary and you add random number, you pick a random number, add to it, give it to him. Now, I could intercept that, but that doesn’t tell me anything, because it’s your salary plus a random number. He gives it to me, he has his. I could intercept both of them, I can’t tell anything. I add my salary and a random number, give it to you, you could intercept all these messages, tells you nothing provably, right? Um, you subtract off your magic number, there’s still three salaries now and two magic numbers, you give it to him, you subtract your magic number, there’s still one magic number in there plus the three salaries, give it to me, when I subtract off my magic number, I now know the sum of the three things. You can intercept every single message, tells you nothing, right? Only I end up knowing the sum of the three things, and I can divide by three and then get the average, right? So, that’s the basic notion of it, and you see that it’s a sort of M squared type of a thing You can do it in a multilevel way, where you get a complexity of N log N. There’s various sort of things you can do With a little bit of tuning, you can get things with essentially order and complexity to them This general idea has been around for for 25, almost 30 years, and I think it was 2012, the method that I just illustrated was shown to be touring complete, so you can compute anything this way, right? Um, and various people had begun using this. For instance, nucleus missiles in the United States use this method to make sure that everybody agrees that the missile should be fired, right? So, I think you can probably trust that it works. It’s hard to crack. Um, we’ve done things, for instance, with NEC NEC has a product, we did this on credit card data, where we combined tellco data and credit card data and government data to be able to look at credit risk You know, normally, pe normally, people don’t share that data, but this doesn’t count legally as sharing The really interesting ing thing here is, first of all, a lot of the problems we have in the world are due to the fact that people can’t share data The bottom one is Larry Lessick, a famous Harvard Law professor, and he has this legal statement about this. It turns out that if you are doing that thing that I just did, that does not count legally as sharing personal data, and the reason is is that I can’t tell anything about your salary from that whole process So, it’s not personal data Moreover, you could be in France, and you could be in Germany, and I could be in the U.S., and we can still do it, even though it’s against the law to move data across borders. So, this begins to answer those questions of autonomous vehicles. How do you do things across borders? How do you do things with data that you don’t own? And this gives you a way to do computation on data that you don’t own without revealing the data. That’s, like, mind-bending. If you haven’t gone there before, it takes you three or four days to sort of figure out that, this is one of these everything I believe is wrong sort of moments. It’s like, wow, that’s pretty weird Um, so, we have a spin-off company that does this, there’s a bunch of spin-off companies that do this, a bunch of the big guys, a little bit of an arms race, so this is enigma.co, not com, because commis a different thing Anyway, um, there’s another way of doing it that has to do with knowing something about the behaviors of the things that you’re observing. So, I’m going to illustrate this with people, but you can do the same thing

with robots. So, if you know something about the programming of the robots and how the robots react to each other, then you can analyze that encrypted data in a way that’s actually dramatically more efficient, and this is a little bit like saying, well, if I knew the rules of physics, I could take observations and fit polynomials to them and exponentials, and I could get really good fits for the performance of something without knowing the, you know, the details of the structure You’re building knowledge into the machine learning. Okay, so the knowledge I’m going to build in here basically is preferential attachment Preferential attachment is, for those who don’t know it, is that if he’s really, really doing well in the stock market, and we can all sort of see that, then we’re going to tend to follow him rather than him, because he’s not doing so well , so we tend to follow the leader, and it’s a rich get richer type of phenomenon, it produces exponential distributions Um, so, if you know that, you can look for these particular, it induces particular statistics on things So, for instance, um, I’ll give a somewhat different example, what we did is we gave out phones to all the people ipin this laboratory and another laboratory, and we asked who is influencing who, who can I predict from the behavior of another person, because we know that when there’s a social relationship, there’s also a physical proximity relationship. You hang out with the people you work with during the day, you hang out with your friends at night, when your boss shows up, everybody else shows up, when you show up, your boss may not show up. Those are very simple sort of association rules that we know are true of humans So, we can look at things like, for example, Bluetooth pings, and we can figure out pretty much the complete social structure from those ten pings Very, very noisy data, in this case, that means nothing, it’s just sort of, um, I don’t have to tell you the physical meaning of it at all, and I can tell you the dependency of those six signals is in that graph, and using that graph then, I can clean up the noise. So, this is sort of a fake example, it’s on the web, you can play with the code, if you want, but this is taking six things that are connected in an unknown way, add a signal to noise ratio of one in the observations of the sensors, and pulling out almost exactly what the signal is by looking at the dependencies between them. So, a more, perhaps, intuitive thing is this, this is from those Bluetooth pings, so what we were able to do is figure out, you know, all of the people, all of the groups, who’s the boss, who’s not the boss, who are friends, who are not friends Just by looking at the conditioning dependency between the symbols that we were getting. In other words, we didn’t need the data, we could have used hashed data, encrypted data, to be able to take, we could have taken pings off of phones, hashed them, and done exactly as well as we did here. So, you have to think about that. Really? Let me give you a couple of examples. This is an example, this is about 10 percent of all the cars in Tokyo, remember, I did this thing for Nissan, so we watched all the Nissan-related topics that use their wings product, and what we found is we could detect, um, essentially trends in traffic better than, you know, Google and Waze and people like that, in fact, rather dramatically better, because what happens is when there’s some sort of problem in the road, like somebody’s digging up the road, people begin making different choices, and they see the car in front of them, and, you know, they’re all responding to this obstacle, and so as you begin to see that sort of thing happen in a non non-coincidental way, you say there must be something going on here, because normally, these things are just, you know, normally distributed, now they’re not, now there’s a co-dependence between them, and as you see a stronger and stronger co-dependence, you say there must really be something wrong there, and it turns out that you can predict traffic problems hours in advance that way

So, it turns out that the Waze algorithm and the Google algorithm are pretty good on average, but they have really big standard deviations in their transit time, and if you integrate data like that, you can reduce it rather dramatically Again, computing on encrypted data, you don’t need to know the actual data to do it So anyway, um, another spin-off, yay, we do that. Indoor.com also works. We have a book that, you know, talks about a lot of the architectures and these, um, these computational methods, I think it’s $12 on Amazon, something like that, not a lot of money, but what I hope, um, is useful to you is, first of all, this sort of federated distributed architecture linked together with blockchain for auditing is, I’ll argue it’s the coming thing. I mean, you know, Europe is buying into it because of GDPR, it gives you the ability to audit and explain what happened, it gives you a much higher trinsic level of security, because the flow of the data is determined by permissions and pre-determined questions, so you know sort of what’s happening, and you can go in and see it So, while it’s not perfect security, right, it, um, avoids a lot of the most common problems, and it gives control, both locally and to the actors, through the permission structure and through the auditing. So, that’s really cool, and then there’s this other thing on top of it, which you probably didn’t know, which is you can, even with encrypted data, you can do a lot of different sort of computations I hope that was useful to you I’ll just end there Okay, thank you (Applause ) >> SPEAKER: I have time for, like, maybe one question or something. Maybe two. I don’t know No? Sir? >> SPEAKER: Um, you confirmed that this is the, you know, algorithm that that particular mode ran to generate that. Um, how do you, so you can have the algorithm, how do you validate that that was the algorithm they ran? >> SPEAKER: So, you caught a problem in all of this. So, you can verify that people didn’t mess with the code, because that’s on the blockchain, but you can’t say, oh, maybe the guy took that code and then sneakily substituted something, but you do have the answers that they generate >> SPEAKER: You can at least reputationally see >> SPEAKER: That’s right. So, over time, you get consistency and reputational If you’re doing the secure multiparty, you have stronger things, because it won’t work unless they behave themselves, right? Yeah? >> SPEAKER: Thanks for a nice presentation My question is, um, how do you think, is it possible right now, or maybe in the nearest future, to create, um, the common format for the security and privacy and the rules for the private data storage and the distributing? How do you think about that? >> SPEAKER: Well — off mic — I think that, you know, there will be a number of different ways to do it in detail, but I suspect that the general architecture will look something like this, just because it gives you those functional capabilities, and it reduces, um, tax surfaces, basically, both for privacy and security >> SPEAKER: Thank you >> SPEAKER: Yeah Speak speak while the next speaker

gets ready, um, let me introduce Dr. Rudovic Ognjen Rudovic is a person that, like, really, um, developed, like, personalized machine learning methods, so he cares a lot about how do we personalize, like, all these methods and algorithms that, like, we heard, like in the AI and machine learning space, and how do we apply them, like, to robots, right? So, we have, like, together, like, a robot chain, which is how to make these robots useful, but also to make them accountable So, without further delay >> SPEAKER: Thank you Um, so, I’ll talk today for the next 20, 25 minutes about personalized machine learning and blockchain for social robotics, and, um, this is a collaborative work with Eduardo and Alex Pentland here in the media lab So, my talk is going to be mainly about intersectional machine learning, social robotics in block, and open algorithms, and applied to the click setting with a particular application in autism theory So, why autism therapy? Those kids have difficulties in social communication with particular individuals, so they undertake a number of therapies which are all types of interventions that help them to improve their social skills, and what it looks like, I’ll show here in this video This is a typical type of social interaction therapy for kids with autism — music playing over voice — to engage the kid and also to help the therapies — music playing over voice — not only to teach the kid these typical expressions, but to monitor the engagement levels of the kid Here, we have different stages The first stage is the learning stage, so these are the typical steps that the therapies will perform on a daily basis for the kids with autism in order to improve their learning outcomes and to make them more, um, capable of interacting in the real world with typical individuals So, one of the main challenges in autism therapy is that, um, it is very difficult to keep the kids engaged. In a few seconds, they can lose engagement, and then there’s no learning. So, the robots provide a very safe environment for them, and kids with autism really like this 3D embodiment of technology, so they become much easier engaged with this type of technology Then, um, this technology has also, um, a lot of, um, benefits for documenting the therapy. What happens, usually, after the therapy, um, the therapist goes back to all these video recordings, it could be hours of data, and tries to look for the signs of the progress of the therapy, so did the child show some relevant cues that could indicate that there was, um, improvement in the child’s social behaviors. The robot can automatically do this by analyzing the behavior of the kids during the therapy and documenting this, thus taking a lot of, um, data from the therapy # therapies So, before I go into the machine learning , I will just show you here, um, how the structures of the data that are needed for the robots to deal with this kind of, um, scenarios, so at the very bottom, we have this that contains the demographics, such as culture, age, gender, medical records about the kids, which

could be the different types of diagnosis, also the medications that the kid is taking, and all these parameters that are know a priori, but do influence the type of intervention which we could like to apply for that particular kid. Then we have interaction data, these are the data that we can collect using various sensors. Today, we have many available sensors, like microphones, cerebral devices, and so on, which can provide all this behavioral data that we call interaction data So, these are the data that’s collected directly during the interaction with the robot, and then at the top, we have the expert data, which in our context of the autism therapy, we call the therapist feedback, and this is, um, these are the ratings provided by the experts, either by watching these video recordings of the kids doing the therapy or providing some other insights into the therapy that can be used to select the general observation and personalized observation for that kid So, one of the very important steps for enabling the robot perception here is to provide the data, so we need to build a supervised model that can tell us from this behavioral data, I would like to predict this certain, um, outcome. For example, that could be engagement level or emotional state of the kid in the therapy. Emotional states are usually encoded in terms of this two-dimensional space, which is valence and arousal, valence saying what is the pleasure of this pleasure level, and arousal showing the alertness of the kid, and, um, in the therapy, usually, this is usually coded by several human experts in order to achieve, um, or to show that this data is reliable. In the case of autism, um, in our most recent studies, we couldn’t achieve higher results than 50 to 55 percent, which is a very low agreement, showing that human therapies are very inconsistent in rating this kind of data So, this is another maturation for us to have an automated system that can provide the estimates of these outcomescomes that are the key parameters Okay, so, the typical pipeline in the human-robot interaction, it applies, of course, to social robots, so the first step is assessing where we combine multiple modalities, and then this, um, these modalities are pre-processed, so this could be visual audiofizz logical data physiological data. All of these are passed through the perception models which are enabled through the machine learning models, in our case, valence, arousal, and engagement levels, and this output uses those parameters to select the optimal interaction strategy at a given moment in the therapy So, the information sensing, now there can be, um, realized using many open source tools This could be the processing tools for the autonomic devices, like for measuring the arousal, um, levels and the heart rate, also the temperature of the skin of the kid, but then we have also computer vision tools that allow us to transfer the gaze, the focal points of the kid, and also the body gestures. So, all this, as you can see, is achieved in an automatic way without any supervision. In addition to these cues, we also can use the audio data This can be extracted using an open smile tool that provides the high levels to help the robot perceive the emotions from speech So, here is one example of, um, a machine learning model that we recently proposed, um,

for robot perception and engagement in autism therapy, and the difference this is different from traditional models that try to fit, um, that uses a one size fits all approach, the data of all kids, the perception of the robot is the same for all the kids. Here, we tried to personalize this perception to each individual kid in the therapy So, what happens at the first layer of this model, which is enabled using the framework of deep learning, I assume that you heard about this, which is a new trend in machine learning where we can learn from big data and then encode a lot of information within the network parameters, and those parameters are embedded through the multiple layers of the network So, in the first layers, we have this process modalities that are, um, fused into other level presentations, and then there is a context contextual layer combined with the medical input, what are the communication skills, so on, of the kid, but also the demographic information, like culture, gender, and individual types So, this allows us to fine-tune this model for each particular kid in the therapy while sharing the information of different kids, and in the case of autism, this is extremely important, because this case shows very diverse behavioral expressions during the therapy, and the model cannot easily generalize all these kids, which, um, the robot perception of the key parameters that we want to measure in the therapy Okay, I just want to show you here, um, what is the benefit of personalizing this robot perception We used 32 kids from two different cultures, Japan and Europe, and we found that in terms of the performance measure, which was inter-class correlation, which was how well the model predicts engagement, um, compared to the human experts by measuring the agreement between these two, and as you can see here on the left, we showed the improvements in terms of that measure of the personalized models versus non-personalized models, which use only one stack of layers without accounting for these individual differences, and as you can see, for most of the kids, we get the significant improvements in terms of the, um, robot perception of performance Here, we also can see that by combining different modalities, visual, audio and physiological, we can improve the robot perception as well However, um, there is a clear benefit in using these models, but it requires the use of personal data The same happens, um, with interaction models of the robot So, the robot receives all this interaction data, then passes them to the perception models, the one that I just showed you is one example of those perception models that can be used here, and then those, the output of those perception models are fed into a very complex interaction model that is designed to implement a certain type of intervention, and depending on the type of intervention we have, we can have many different models here, and those models are also constantly updated and informed by the background data, so the therapies provides the feedback, how was, um, the progress of the kid during that therapy, and then all that information is embedded in the interaction model, and as you can see, this requires a constant access to the personal data, which is something that, um, we did one hospital we did one hospital, but happens when we have a network of hospitals. Having a centralized database may be challenging, in case when we have the data distributed in different hospitals, we call them here hubs, and also, um, not only the three will need to send all the data to a

centralized place, but also, this data could contain a lot of personal information, which in many cases, especially when working with, um, with kids with autism, sharing the personal data, like medical records and, um, also the medical history, the audiovisual information, it’s not very good, but also, we don’t want to compromise this data. So, we are interested in going from this centralized system to a decentralized learning system that doesn’t have the data stored in one single place, and with the system that we propose here, we can achieve it without sharing any personalized data. So, the system we promote is called robot chain, and it uses the concepts of blockchain, open algorithms, and federated learning, and I will just show you now, um, the key steps that implement this, um, paradigm, and I will show you that on a very high level, but we will have time for questions after I finish this So, the first step is the local learning of the robots So, the robot needs to access certain data about the kid in order to perform the intervention, and to do that, based on the background data of the kid, like demographics and the previous experience from the therapy, the robot will like to send a query to a data server that contains the data from the trusted parties, and this will be, like, from different doctors, educators, public institutions, which is distributed in different places, and the robot cannot have access, or is not permitted to have access to all this raw data, but using the concept of open algorithms, we can use, um, these algorithms that are, um, validated by human experts, and they allow us, they allow the robot to send certain queries that would help the robot to get the data in type of the aggregated information. It doesn’t reveal any individual data about other kids that may have a similar condition. So, the robot can access the aggregated information and use that information to improve its interaction model So, once the robot has, um, obtained this data in form of the answers, in step three, what happens, if you want to include level of security here and allow for a later auditing of this service, we can store this in sections by the robot on our local blockchain, and that blockchain can hold only the first value, but the actual QA answers can be stored, um, locally This allows the expert to see if the robot was actually querying valid and ethical questions about this kid when forming its interaction with them You may ask why do we need this here, why we don’t use any simpler aggregated information that can be obtained without these open algorithms. If it happens that the robot is compromised for any reason, it’s stolen from the hospital or something, someone may access all this information from trusted parties, and we want to prevent that, so by using this system, we put many constraints on accessing this trusted-party data and prevent misuse of the personalized data. The final step in the local learning of the robot is, um, once the interaction is finished, the robot is arranged to be the new interaction data, which is, um, video, audio, and physiological data collected from this kid, and that can be stored locally and used to, together with the therapist’s feedback on this data, to update the machine learning models that are used for the perception and the interaction with this kid. So, I’m slowly increasing the complexity of the system, so now, we have this robot that will, that

interacted locally, so now, we are seeing one hospital, which we call a hub, that contains multiple robots. So, each of these robots has access to this local repository, from there, it can check the latest model, but also, after updating this model, it can commit a new model to this repository, and then that model can be evaluated locally within the hospital by different robots, and then if these different robots also confirm that this new model is more effective for the therapy setting or the interventions we want to implement, what happens is that that model becomes a new candidate model that is going to be shared with the other hubs, which is other hospitals Finally, um, we arrive at the step where we, um, deploy federated learning. So, the source robot has published the candidate model, the model is publishing hyperparameters or the deeper networks that are used to implement the interaction and perception models, and, um, these hyperparameters are now available for download by the other hubs from the trusted hospitals, and then those hospitals, the robots from those hospitals can use that new candidate model to evaluate it in their legal setting and assign a certain score, um, to this candidate model Once, um, a majority of these robots from the network have evaluated the model, we can see if the consensus has been achieved, that this new model is more effective than the previous models deployed in these hospitals, and if this is the case, um, we accept this model as a new base model for the targeted intervention that we want to use for particular autism therapy. What is important here is that there’s no need to store personal data of the kids, so the robot only publishes the hyperparameters, the personalized data that are used to update these models stay, um, at the local hub, and finally, so what we have here is that, um, once the model has been accepted, the source robot is required to publish this section on a blockchain, so we can, for this, um, we can use any type of a blockchain, it could be a private blockchain, where only the network of the hospitals has access to it, but also, it could be, um, a public blockchain that is accessible by trusted parties Um, however, because those actions have been recorded only in the form of, um, what happens is this blockchain cells will contain the timestamp of this transaction, the hashed strength, and the data, but this data will be encrypted in terms of the, um, containing the parameters, the hyperparameters of the model that is accepted as a new base model Another way to implement the sharing of these models is through the use of smart contracts, and it’s something that we are investigating now while we are also trying to, um, deploy the system in hospitals that are performing different autism therapies I will just summarize briefly So, from the work we’ve done so far, we’ve found that they have a great potential to improve today’s, um, healthcare interventions by assisting the therapists and helping them off-load many of the work that needs to be done, but also to help to engage the kids better within these therapy settings, which is important for the learning outcomes of the kids. Personalized perception interaction of social robots, um, it’s very important, if you want to achieve, um, higher performance within the specific settings and for specific kids, especially because kids with autism have very diverse, um, expressions of their behaviors, so for the standard robot perception models that are maybe applicable to typically-developing kids, um, in this context, it is very challenging because of these highly pronounced differences, and finally, um, blockchain and

federated learning are very powerful concepts when it comes to data privacy and data sharing, which is extremely important when working, again, in the healthcare applications, and in our case, autism therapy, where the data of the kids with autism needs to be saved securely and by no means can be compromised. Okay, I would like to thank you (Applause.) >> SPEAKER: If you have any questions, I think we still have 5 minutes now >> SPEAKER: I’m still not sure I see why a blockchain is needed as opposed to just a limited access database >> SPEAKER: Yep So, the blockchain, if you use a blockchain, it’s an efficient means just for recording these transactions, because it allows also other parties to get information about, um, the most recent transactions performed within the local network So, I it could also be used for limiting access to the database This is just a more advanced way of using it at the moment >> SPEAKER: I completely agree, but, like, here, like in this slide, you see, like, two types of systems, so you have, like, a private critical network, in which robots actually could share these kind of, like, reposts, these kind of, like, hyperparameter models. These, it’s okay, like, if we don’t use, like, a blockchain, as long as we maintain this trust within the system. So, you authorize the fact that you are a hospital, you are a trusted party, but the problem is that, um, once, like, you get these robots talking to each other, you need, like, a way in order to understand the concepts of this model is accepted by everybody, and everybody, like, played according to the rulesof, like, the internal network. So, what we do here is we have, like, a public blockchain, in which once the robots are getting to consensus about the new candidate model, you just publish a digest of that, so you get, like, the accountability, you get, like, a station that that happened at some point in time, so you can understand that no single party within the clinical, the private clinical network, um, deceived, like, the others or played with the others >> SPEAKER: Yeah,gist like to add, yes, we can sort of the history of these transactions, but, um, private databases is also one possibility here >> SPEAKER: More questions? >> SPEAKER: Um, I have a question Please tell us what information you’re putting on the blockchain, because I get you’re uploading not the whole data — >> SPEAKER: No, here, I’m uploading the timestamp, of course, of the transaction and what allows other networks to access the specific set of hyperparameters, which are stored also in the O chain in the form of encrypted data, but this doesn’t take much space, because these are only the hyperparameters of the network, so it’s a very low baseline, which wouldn’t consume a lot of memory >> SPEAKER: You mean that you’re encrypting the data, putting in the network for the access to the other participants, and they have, um, keys to encrypt it? >> SPEAKER: Exactly. All the local networks have the key to these parameters >> SPEAKER: Okay, and what about key security? >> SPEAKER: I can leave that, again, to Eduardo, because he works on that part >> SPEAKER: Yeah, deaft definitely. Right now, we have a working simulation, so that part is being controlled, so we generate new keys which cannot be, like, shared outside, like, a private network, right, but, like, now that we are entering into, like, the situation in which we’ll have to deploy the system into real hospitals, we’ll have, like, to seriously care about, like, how do we generate these keys and how do we maintain the security of that. So, it’s working progress. Right now, we’re moving from a

simulation phase to a deployment-like phase >> SPEAKER: If I can just build on that, um, specifically because you mentioned that it’s a public, um, blockchain, um, so, Ed Fe Felton since 2013 was saying that because of the pseudonymity of blockchain and also the linkability issues on blockchain, this is definitely not going to be incredibly private data, and there’s an upcoming book, I think you might be interested in this, um, Michelle Fink wrote a book on the GDPR and blockchain, and there are definitely some tensions there, so I, perhaps, would like to challenge you to reflect on some of these tensions >> SPEAKER: Definitely Thank you very much >> SPEAKER: Any other questions? No? Well, if there’s no further questions, let’s thank the speaker and welcome the next one (Applause ) >> SPEAKER: Well, while the next speaker prepares, so, I will introduce him. We’re having here Dr Thomas Hardjonoo. He’s responsible for us being here, because he was the guy that convinced me that there was, like, a possibility of, like, mixing robotic systems with blockchain once I arrived here to MIT So So, he was the private director of the MIT consortium, and he’s going to talk about, like, blockchain interoperability, so how can we link, like, these different blockchains in order to work as a whole system, and also how do we pass, like, from a data-based system into a transaction-based system >> SPEAKER: So, folks, while he sets up, just want to say thank you for coming to MIT It’s a beautiful day, I almost wish we were out there, enjoying the sun, but it is, what is it, minus 6 Celsius, for those who think in Celsius, at least that’s what my mirror said this morning, I said, oh my goodness, minus 6, I hope the car starts, but, yeah, welcome to MIT. We have an exciting, thriving group at Connection Science, where we’re a group of people working on trade coin, blockchain under that umbrella, a project called trade coin. We have economists, we have people interested in supply chain management, how to use in-chain sort of tokens to do cell window supply chains, which is an exciting topic, we have people like me working on infrastructure security, so that’s what I’m going to talk about today, is infrastructure security and interoperability, which I believe is going to be crucial for anybody deploying blockchains. So, the previous diagram, when you have two independent blockchains, and if one’s private permission and one isn’t, how are things going to work out, right? So, that’s what I’m going to talk about, but anyway, I’m going to skip slides back and forth, because some of this is going to be pretty boring, but, um, for some of you. I’m going to show this, you know, almost mandatory slide, you know, every time I think blockchain technology is about to go over the hump, someone quickly makes up, puts a verb and adjective behind it and quickly rolls the ball back again to the left side, so we’re never, ever, ever going to get over the hump, you know, we’ll never get to the plateau of product productivity I’m kind of skeptical, cynical, but anyway, we’re in a good time. Unfortunately, I was a young engineer in the 90s, when the first Internet revolution occurred, and it was just as messy There was as much BS coming out as we have today from

conferences like Consensus, all these cocky guys up there on stage, saying they’re going to revolutionize the world. I’ve seen that before, people are saying we’re going to have IT television all the way into the home I don’t know, do we have IP television? I have Netflix. Is that IP television? So, we can argue fine points, but I’m going to talk about this thing called interoperability, which is actually very crucial and actually very important historically. For those who are interested, for those who weren’t around, in the 80s, there was a huge debate about, um, LAN network, IP network interoperability So, in the 80s, there was there were at least three big networks out there , for those who know DEC, Digital Equipment Corporation, IBM, and then a whole slew of these LAN vendors, and they did not talk to each other on purpose, and this is like today Everybody wants to come up with their own blockchain, and no one’s really talking about interoperability, but people aren’t realizing interoperability is so crucial to survivability So, um, in this mess in the 70s and 80s, DARPA basically issued, like, a grand call, a call for grants, saying we want to create networks, but here are the goals, and, so, they came up with seven goals, three are fundamental, non-negotiable, those other four are, like, yeah, you know, if you get to it So, the three goals, in doubt, communications must continue end to end, regardless of the loss of networks, entire networks, and the loss of gateways. This was number one goal, and there was a motivation behind this These are, you could say military values, there was no, we were not trying to hide this, the fact that DARPA said we need this for survivability of communications in the United States. The public networks, the private sort of corporate, non-DOD networks must be able to have interoperability so that, again, this is, we’re talking about the Cold War, we’re talking ability about the U.S just finished Vietnam, Russia was our number one enemy, if Russians were to knock out the military communications in the United States n U.S D, the U.S. DOD must be able to take a router out of their DOD network, plug it into their commercial network, and it should just operate, it should just work So, it’s the same gear, same protocol, maybe different cryptokeys, but that’s a separate thing. So, that was the motivation. Secondly, it needed to support a variety of service types So, um, if you do FTP versus you do TCP connection, how many people still use FTP? Actually, your browser does, so it should behave the same way, it should support that type of service and variety of networks It should be able to traverse, whether it’s DEC net or IBM token ring, it should just work, and token ring outperformsforms ether net any day, for those who know LAN history. We love IBM, but they always overprice their hardware, so they sort of outmarket themselves by mistake, but that’s a separate story So, let me skip through this and come back to It so it So, the question today, can our, um, you know, blockchain infrastructure withstand attacks? So, there’s a lot of attacks that can happen, the classic network attacks, there is going to be a family of consensus algorithm manipulation attacks, it’s no doubt, and she’s nodding, because she probably knows the literature, and the guys at Cornell, they’ve actually looked at, you know, they’ve studied the performance of bit coin, for example, under different constraints, and it’s very clear you can manipulate the consensus, the protocol, pretty easily. What if you recognize legitimate applications? I mean, crypto keys, that’s, like, the most awesome attack tool, right? People know what I’m talking about

If you want to grind down ethereium, you know, hey, you do another crypto key, right? This is, like, that perfect tool, and what about, you know, viruses for mining software? The minors out there, they don’t exactly have pristine error-free code, you know, it’s just a matter of time for something like that to happen. So, what are the lessons learned? So, um, interoperability across networks is fundamental for the survival of a whole. So, let me put it in today’s language If you have an asset, tokennized asset sittingen sitting on a blockchain, and the blockchain is under attack, it’s not responding, you sent a transaction, it’s sitting there, you’re not getting any confirmation, you say I want a movement asset, that blockchain is not performing so well, guess what? You can’t You’re stuck, and maybe the goal of the attack is, in fact, just to cripple the blockchain so that no one can move assets So, this is reminiscent of controlling a network so that all the IP traffic can never go through it, right? Exactly the same situation as back in the 70s and 80s The other lesson learned is, um, autonomous systems, so there’s a whole debate, for those who know what the IETF is, it’s the body that defined all the standards for TCPIP, IP routing, goodness, basically everything, thousands of RSCs, and I MIT has been there since the early days days, I remember him up on stage, leading the debate about autonomous systems, but a network is an autonomous system So, I don’t know how many people know, we don’t have a single Internet, there is no such thing as one Internet, we have islands of networks that are stitched together, and your packets, you sent a packet to your e-mail, it’s going to traverse through different networks throughout the United States, right? And the reason why that works is that each individual network is owned and operated by some legal entity, they actually have a registration process to do that, and they have peering contractual agreements with one another that are standardized, but there is a fixed boundary, there is a fixed ownership, and it’s, every network is treated as autonomous, they must be able to control their own performance within the network, they don’t need, necessarily, to expose the routing condition within a network to other ISPs who may be competitors So, why I put this up is is the blockchain system an autonomous system? So, take bit coin, are there edges to bit coin? I would say, yeah, the mining nodes could be the edges. The population might increase and decrease, that’s an interesting question, do we want that? Our synonymous or atonmous mining nodes, is that a desirable feature? Some might say yes, some no. I would say for certain applications, yes, for others, you know, not sure So, in the Internet, how many people know who Vince Surf is and why people make a big deal? Shows your age (Laughing.) >> SPEAKER: Sorry, and why he is such a big deal and why he is reversed? So, maybe, in today’s language, he could be, I would say a bit like metallic, but I think more, because this guy, um, representing the United States with the OSI to standardize a thing called the data ground, right, and this is what it looks like, right? And this is the famous paper there, if you want to look it up, May 1974, who knows who Bachan is? What’s his other contribution to the world that you guys in academia use almost every day? Who knows what the DOI is? That’s him, and he was e-mailing me last time saying, gee, these blockchain guys, aren’t they just reinventing the DOI? But this is what the data gram looks like, type of service, total length, source address, destination address, does that ring a bell? It’s pretty interesting, right? What else could you load there? So, options, so, this is the header. The body of an IP

address is flexible, it’s a type link value structure , we can sort of simulate the same thing for blockchain transactions. So, the question I throw at different communities is why don’t we have something like this for blockchain transactions? Where there’s the same transaction format, I can feed it into an ethereium network, a bit coin network, a hyperledger notework, literally, it should just work. I don’t get how these guys do consensus algorithms, literally, I don’t care, I just want to know my transaction will work on N number of blockchain systems So, gateway, so this is why, what are gateways? So, um, when these ISPs were just forming as companies, corporations, in the 80s, they realized that in order to do peering and transfer of packets across networks, they needed high performance, what are called big iron routers, which are called gateways. You can buy them today, it’ll cost you at least maybe a couple hundred thousand dollars. These are, like, huge boxes, right, and they have them today, but the idea is that you would actually, um, peer them across ISPs So, an ISP would have maybe six to ten to two dozen of these boxes, and the other side would have the same, and they would be able to route packets through, and the gateways would just pair with each other, depending on traffic condition So, could we use the same gateways model for blockchain systems, right? So, in order for autonomy to be maintained within one blockchain system A and blockchain system B, they could, assuming they have the same sort of data gram format, they could run different consensus algorithms, don’t care, but what if these, each of these blockchains could have gateways that are intended to transfer transactions across, so that if I want to move my asset from blockchain, it could be, say, my house registry, you know, I’ve got LAN somewhere, and you know what, I’m going to sell to Eddie, Eddie’s a buyer in China, and he says that’s great, but I want it to be honed in my blockchain system in China, which is a legitimate request, so how do I actually move this asset thing from blockchain A to blockchain B? We need a thing called gateways So, the question then becomes, um, the mining nodes are the gateways Number of interesting questions What are gateways? Those special boxes? Are those, um, mining nodes that have been upgraded to special status for a particular set of transactions for a particular amount of time? Do they need to have specific types of secure hardware? What is it? So, these are the issues that we’re studying So, we came up with the definition of interoperable blockchain architecture, the same way that the DARPA guys defined what the Internet was There is, in fact, a definition about what the Internet is It’s a paragraph long. So, and I’ll pass through this slowly So, we need to be able to distinguish one blockchain system to another So, my instance of hyperledger code running with a bunch of nodes here must be distinguishable from his bunch of nodes over there, and that’s easy, but what if it’s, you know, one bit coin instance there, one bit coin instance over there? Um, autonomic transactions may span multiple blockchain systems So, what if I was doing a transaction where, in fact, if I look at my transaction level, it had three, it was move asset, send money, you know, um, whatever register new asset in some new registry data? Those are three different transactions, and you would consider the application-level transactions to be completed, if all three had been completed altogether So, the whole thing needed to be autoomic, and the problem is you’re talking about an autoomic transaction that may span different blockchain systems, and we don’t know the performance of these, we don’t even know if they talk to each other, right? Secondly, once a transaction is recorded, how do you ensure reachability? So, I’ve just moved my asset to a new blockchain, I’m in the process to do that, the new blockchain system, the party, the applications running, how does it find out in my home blockchain system that I’m actually the legitimate owner of

the asset, I actually have the private key? What if my blockchain system was a permissioned system? How do people outside, you know, read/write and do all of this in a compatible manner? So, this is a definition, but it’s also a list of problems You could do a couple of Ph Ds on some of these topics So, the other question is algorithmic trust, so we’re working with some folks at Intel on this, and, so, I don’t like the word trustless, because the bit coin mining nodes are not trustless. So, let me rephrase, for those who are not deep in the bit coin model, in bit coin, each mining node has the independence and operates independently in doing the proof of work, as long as they know what the transaction is They grab a sample, they try to do this hash matching, and they’re doing that independently, right, without any communications whatsoever So, the only thing that everybody agrees on in that network is that everybody needs to use the same hash function, whatever it is , and everybody relies on the fact there is knowledge that the shock algorithm has not been broken. So, we are standing, we are literally standing on a belief that a mathematical proposition has not been broken We’re standing on pure cryptography. I don’t know, maybe some guys in Virginia have, you know, knowledge of how to break a shock function, you know, given so much money on bit coin, I would not want to advertise that, I’d try to cash in, right? I don’t know, but looking at gateways, is that model sufficient? Can you get a bunch of gateways belonging to this community that’s defined as a blockchain to operate independently, or do you need other things? We think it needs to be assisted with existing forms of trust So, what we mean by institutional trust is that when you buy specific types of hardware, say from Intel, from Gemalto, the people who make smart cars, we are standing on the reputation of the company, and we are standing on some form of legal sort of service contract with that company, so it’s not completely trusting the unbreakability of crypto algorithms, it’s actually standing on some legal jurisdiction in combination with company reputation and so on and so forth, and as I was saying before, how do we make sure that gateways can talk to each other? If two blockchain systems are so, so different, could we at least guarantee that the gateways have some commonality, that they can talk to one another, right? So, these are, it’s a quick diagram, just to show this, application X and application Y Through some selection process, again, TBD, G1 and G2 decide that they’re going to, um, be peering with each other for this particular transaction, and in this, as I mentioned in a previous example, I’m trying to, application X is trying to move an asset from, um, the ledger in blockchain BC1 to blockchain BC2, right? And, so, is there some kind of a hand-off process where, um, it’s almost like, you guys know what a two-phase commit algorithm is? It looks something like, it may apply to this. So, basically, G1 and G2 has to communicate so that G1 marks the ledger in blockchain A, saying, okay, this asset is about to be moved, but not confirmed yet, and it talks with G2, and G2 begins a pre-commit on the other ledger, saying, okay, we’re about to receive a new asset in our, um, there’s a new transaction that’s going to be submitted to the ledger, and only after the transaction in ledger BC2 has been confirmed, that it’s going to be released and marked as permanent on ledger one. So, that’s the kind of basic model We’re still working on this, but this is reminiscent on how IP gateways, um, operate So, some open challenges, how to define the perimeter of a blockchain autonomous system, and I call it that specifically If you agree, it’s an autonomous system, what is the boundary?

Can a mining node be a member of two blockchain systems? In this sense, it’s dual-hitted. It’s running bit coin software, then it’s running hyperledger. Or are there some applications, you know, financial applications that exclude that possibility? Do we need to know the identity of, the device identity of mining nodes? Do we need to know the legal identity of the owner of mining nodes? Yes? No? I don’t know, but I’m pretty sure the computers over in Wall Street and naz and nazdack are not anonymous What is the minimal assumption? It is the data gram. We assume all the networks operate on a same transaction unit definition How to interoperate across permission systems, so this is delegation and trust, this is, this topic has been a big deal in computer security over two decades, three decades, and it’s not getting easier. What is the business model for peering? The ISP model, the idea was that the customer who was attached to the, um, ISP directly would be paying to just subscribe, like today, Xfinity and Comcast, they are an ISP, but when an ISP one peers an ISP two, they enter an agreement based on the volume of traffic, and they play tricks on each other, by the way. This is why we have legal agreements If an upstream ISP knows, um, that they’re about to be hit with, you know, a new Netflix movie or the Victoria Secret video thing, they’re going to have a whole increase in volume, they will try to deflect incoming traffic from an adjacent ISP on purpose, like literally dropping packets, right? And suddenly, they say why is my backup not going through? So, ISPs play tricks, and I believe the same kind of tricks are going to happen in blockchain systems, because, you know, it’s performance, and performance has a direct, you know, correlation to basically income to the entity of the group of people in that blockchain system So, some key takeaways, designing for survivability is designing for scale. The Internet scales today, it’s not by accident, it’s because they address specifically the survivability question, which leads to the interoperability question, which led to the creation of data grams. There’s a direct correlation. If we do not have the equivalent of a data gram for blockchain systems, we will never have scalability. So, I hear people talk about, oh, the future is going to be one singular permissionless blockchain system out there, I say, I don’t know, who’s going to pay for that? Interoperability is crucial for survivability. Blockchain systems are autonomous systems, we just haven’t accepted it yet, but I think they are Infrastructure components must, may need to be identifiable and ought authenticable. That’s not an English word, I just made it up Um, we need some degree of technical trust, and technical trust means trust, um, established through the hardware and the creators and the providers of trusted hardware, and we need peering models, contracts, paper contracts, not smart contracts, just between two communities, and that’s the end of my slide Let’s just have a discussion, Q & A. I think I have time for that, right? (Applause ) >> SPEAKER: Any questions? >> SPEAKER: Any questions for the speak speaker? >> SPEAKER: Thanks so much for your presentation. Really, real really exciting. I just want to, again, challenge the idea that, um, basically the blockchain system is actually autonomous So, there’s also this book, um, she has, she gives a lot of attention to the Dow, which I think by all definitions, can be considered to be autonomous However, for the rest of the activity on ethereium, for instance, can you really argue that a ledger, like the one that ethereium uses, is really autonomous? Um, I mean, even the hard fork around the Dow showed that there is still a lot of human agency involved in how the network operates, and, so,

where does this human agency end, in your view? (Off mic ) >> SPEAKER: Going back to the Internet model, the autonomous system definition was created because of technical necessity and legal necessity Technical necessity meant you had to have a bounding to your router domain. You have a bunch of routers, and they have to operate at certain performance, somebody needed to own and operate them and feed electricity and plug in the cables, and so it was a legal company. Is ethereium like that? Who owns ethereium? Okay, when you pay for gas, who makes money when you have to pay gas in ethereium? Should I be rude or not? So, the question is, it’s almost like, implicitly, maybe they’re effectively the legal owners, it’s just there’s no legal paperwork underlying it They’re making money. Are they paying taxes? I don’t know. If they’re paying taxes, maybe they’re legal entities, right? So, maybe, what we need is a more, a formalization of the notion of ownership and responsibility So, when crypto keys occurred and my transaction was slowed down, was there a support number I could call up and complain to? Was anybody held, you know, responsible for this? So, the question is for very high value financial transactions, for example, that kind of model may not be acceptable, right? It’s fun for people to talk about trustless and independence and autonomy, but if you’re losing money — (Laughing.) >> SPEAKER: You’re not happy, and at the end of the day, when individuals, companies, lose money, you know, things go, systems become up unacceptable, right? So, I don’t know if I explained that to you We’re not there yet, and at MIT, we’re saying let’s cut to the chase and say, look, autonomous, blockchains are autonomous systems, but the definition might be different What’s the definition of an autonomous blockchain system? For example, one definition could be all the mining nodes need to be identifiable, we can’t have a situation where the majority of miners are actually owned overseas in Asia Is that acceptable? So, if this blockchain is relevant, if blockchain systems are going to be so relevant to national economy, do nations have the right to question the ownership and responsibility, just the way the NASDAQ is responsible to whoever the body is, right? So, these are real-world issues This is where technology meets reality, technology meets economy. So, sometimes, engineers, including myself, we kind of dream too much, but, you know, this is, I don’t know if that helps, but we’re getting there >> SPEAKER: Um, I have another question I was curious if you could say, um, what you think about projects that work into the direction of interoperability. I have in mind polka dot. I don’t know if you’re familiar or not, or it doesn’t have to be about this specific project, but in general, do you see attempts, um, that are coming out towards interoperability, and how would you situate them in relationship to what you think is necessary? >> SPEAKER: The two levels of inter operability, what I just talked about is what, I would say the mechanical level, everybody agreeing on the same transaction data format, just that. So, if all these, um, sort of leaders of blockchain communities could agree on that, that would be superb. Today, when I hear people talk about compatibility, you know, interoperability, they really mean ERC20 compatibility So, if people know what ERC20 tokens are, so, so long as a blockchain says I’m ERC20-compliant, and the other side also says, yes, I’m ERC20-compliant, things can flow, for sure, but that doesn’t solve the underlying transactional compatibility level So, I’m not, I don’t know the details of, um, polka dot There’s also efforts, people doing interoperability of wallets. Maybe there’s a role there, but we’ll see, but I don’t know if that kind of helps We need to talk about two levels, not just one, not just at the value, at the token format level, because that’s pretty much, maybe the whole world needs to just adopt ERC20

The question becomes how do you bind the value of the ERC20 to real-world value as perceived by human beings and the rest of the economy >> SPEAKER: Thanks >> SPEAKER: I wanted to know, given the, um, especially with permissionless blockchains, like the multiple iterations of bit coin now, um, given that it’s so based on an incentive structure, I was wondering if you could maybe talk about incentives for those kinds of blockchains to even try and get into interoperability, especially as it regards to asset-shifting, because, you know, one of the features of bit coin is that there’s only ever going to be 21 million of the darn things >> SPEAKER: So, I think the trigger for people wanting to get together and talk is when everybody hits a wall in terms of, um, performance and customer satisfaction. So, I don’t think we’re there yet, because people are not really as yet moving assets across and so on, unlike the Internet, from day one, the goal was to send an IP packet from MIT all the way to, you know, Berkeley in just one packet The other thing I complain about, guys here know, this whole hype and speculative investment in blockchain is actually a distraction and a disservice to, like, research that needs to be done for blockchain systems, right? Again, essentially, DARPA was dangling a huge check, and everybody said, yeah, that’s fine, and even in the 90s, engineers, they just wanted to find the best solution, but back to your question, we’re not there yet, but, maybe, you know, if one day, we have a few crypto keys out there operating and things halt, um, and, maybe, the thing might be national level, you know, a whole bunch of blockchains in China and a whole bunch of blockchains in the U.S , and they don’t interoperate, investors start to complain, why are engineers not making this work with one another, maybe that’s going to be sort of the turning point >> SPEAKER: Thank you very much, Thomas (Applause ) >> SPEAKER: Okay, I’m going to introduce the next presenter It’s Professor Stepahine Jill Stepahine Gil She spent some time here at MIT, actually at the other side of the street, so she’s been doing, like, ground-breaking work about robotics and security, especially in the field of, like, multi-robot systems, so I think that, like, what she’s going to show, like, today is, like, ground-breaking work I hope you enjoy it >> SPEAKER: Thank you so much So, it’s really great to be here. As Eduardo said, I spent some time at MIT, I got my Ph D across the way, and I’ve been at Arizona State University for about a year now, and I’m excited to tell you guys what I work on. So, I work on security for robot systems. This is a little bit different from blockchain. I don’t work on blockchain per se, but I think there is a lot of synergy there, so excited to talk to you guys about that So, when we think about security, this is, um, data systems, right? So, makes sense, we understand what security means for data systems, but what about these systems, multi-robot systems? What does that actually mean? Um, and that question is going to become more and more important, because multi-robot systems are not going to be secluded to these environments, such as manufacturing, factory environments, where we don’t really think about them in our every-day lives. Multi-robot systems are becoming much more pervasive now and in the near future, both in our house and on our streets So, actually, in Tempe, in Arizona, there are a lot of autonomous cars that are on the streets already today So, just, um, looking at two very, um, large examples, self-driving vehicles, Forbes predicts that a conservative prediction is that there will be 10 million self-driving cars by 2020, and the top benefits for self-driving cars are often quoted to be things like safety or efficiency. In fact, the national highway and traffic safety administration said that Americans spent an estimated $6.9 billion in traffic delays So, one of the really kind of nice ideas of self-driving cars

is that maybe people will be able to get around much more safely and efficiently. So, some of this work is actually from MIT, the sensible city lab, and people have been thinking a lot about designing algorithms, of what these multi-robot systems would actually look like in the real world. So, another example are drones So, you guys have probably seen these trends, that drones are becoming much more present In fact, research markets estimates a large growth, um, over the next few years, and Business Insider also has these exponential growth graphs for what we’re going to see in terms of drones in the near future, and one of the main application areas for drones are delivery services, and Amazon has been talking about this for awhile, but actually, China has already approved the first drone delivery services for meals, and there’s an estimated reduction in last-mile delivery costs of about 80 percent, so we’re talking about very big savings, especially when you’re thinking about companies the size of Amazon So, hear here here’s a concept of the Amazon Prime air drone, and here’s also a concept video of what these systems might look like So, here, you have a multi-robot swarm of drones that essentially aggregate at a certain area, pick up packages, etc., and then start transporting those packages through some future urban environment. So, maybe this is what delivery will look like in the future, and, so, all of these are multi-robot systems. What about security for these systems? We’ve been really focusing in the past few years on making these systems more and more capable. If these systems were to come into our every-day lives, what would they look like? How would they help us? How would they get us on the roads more efficiently, more safely? How would they deliver our packages? But what about actually securing these systems? So now, I hope that you guys start getting a sense for the importance of starting to think about these kinds of problems, and as Eduardo said, when he first started, um, when he first gave the introduction to this symposium, these are questions that are just now starting to be asked. So, the question of security for traditional network systems, I think we understand, but for these other types of robot systems, which, by the way, a robot is just any machine that senses the environment, takes that information, and runs algorithms that make decisions on that information and acts in the physical world, so these robot systems have a lot of the same security concerns that you would have in a regular network system, computer system, Internet system, but in addition, they take action in the physical world So, securing these systems is arguably, um, an even more critical need. So, just to kind of give you an idea, researchers have already started trying to find holes in these systems For example, UT Austin showed that an $80 million yacht could be, the GPS system could be spoofed and actually drive the yacht off-course, and then, also traffic control, so there’s been some research in spoofing traffic control GPS, and also spoofing regular, um, what you’re familiar with, Google Maps, but these cases are not isolated to research actually So, you guys may or may not have heard about this, it made the news, so this was a funny story actually of a guy who was driving along the highway near the Newark Airport and had a GPS jammer in his car, because he didn’t want to be tracked by his company, and that actually disrupted Newark Airport. They had to actually disrupt their operations, because their navigation systems were getting spoofed, and so this person actually ended up getting a fine of $32,000 for this, and these kinds of jammers, the kind that he had in his car are very available, and these are human-operated vehicles. When we start thinking about autonomous robotic vehicles, these kind of spoofing of data becomes very important. Here’s another example So, DARPA, the defense agency, has been doing research into whether it’s possible to remotely hack a

car, and that’s what they precisely did here. So, they showed that hackers could remotely access the controls of a vehicle, making it such that the human driver could not operate the car. So, how does this actually happen? How do robots get hacked? Well, you can think of these as two different types of systems. So, a car, a single car, like the DARPA one that we showed in the previous video, these are a group of sensors that are all communicating information to some kind of a central controller, and any one of these communication links can be, um, accessed, for example, and if they are accessed, then the information from those sensors can actually be spoofed or hacked Another example is a multi-car system, where now, these cars are perhaps communicating with each other, for example n red car might communicate with the blue car to say, hey, I’m coming up to the intersection, is anybody there, is it safe to pass, or might communicate with, say, some static infrastructure in the intersection, such as a camera or a traffic light. So, again, in this case, any of these communication links can be, um, can be accessed and can be used to insert sporadic data. So, one question is should we get rid of communication? Because that would make our systems more secure, but that would really defeat the purpose and would really limit the, um, impact of these kinds of systems, because, for example, let’s think of autonomous cars, if autonomous cars could communicate with each other, then they can democratize sensing, they can understand what dangers there might be around corners, and in that way, achieve this idea of safer driverring than what a human could achieve, because we can’t see around corners, for example. So, the idea is not to get rid of communication, because that’s really the strength of these systems, so that means that we’re stuck with an inevitable problem, that this vulnerability is inevitable, so we just have to think of new ways to secure these systems. So, if we compare these systems to traditional computer networks, there are a lot of very, um, unique challenges to these systems, such as they’re highly mobile, highly dynamic, there’s almost never a centralized decision-making unit, so highly distributed, and there might already be some compromised agents inside of the network that you have to worry about So, traditional security methods may not be a complete solution to these problems, and so what I argue in my, um, research is that while these systems, these physical networks bring new challenges for security, they also bring new opportunities to rethink the way that we actually approach security, and in particular, what I look at is in physical networks, now you have mobility that’s sometimes controlled, and you have communication, so these agents, these robots are almost always communicating with each other in order to coordinate, and that’s something that we don’t necessarily have, those physical channels of communication, wireless signals in, for example, static networks or Internet networks. So, the question that I ask in my research is can we exploit that somehow to get more information from the system and use the communication itself for an added layer of security, not just keys or cryptographic, um, approaches or afontication, like what you would normally see for other types of networks, and so the main key idea here is can we use the physics of the wireless signals to actually secure these systems, and so what my work focuses on is precisely this question; securing multi-robot coordination tasks by taking a new look at communication, which is something that we already have in these systems. So, there are a few different multi-robot coordination tasks that are very important and that I’ve looked at in my research. One of them is coverage. That’s the example of where you have robots that are perhaps doing drone delivery, and they have to cover an environment, say they need to deploy over a city landscape in order to be able to deliver packages or, maybe, um, have some security functions, etc. Um, drone delivery, which is, in some

ways, related to coverage, but might be talking more about the last mile logistics of how to actually do the deliveries, and then there’s consensus, and we’ve talked a bit about consensus already today Consensus is one of the most important, um, objectives of a multi-robot network, and, essentially, consensus is whenever you need these robots to agree on some quantity, and so I’m going to focus on the work that I’ve done in particular consensus because of its importance for multi-robot systems So, there are different types of consensus that we might need to actually be able to achieve in these kinds of systems For example, consensus on velocity. So, going back to this intersection problem, the way that these problems work is that if you look at these videos, it’s pretty clear, it’s a little bit jittery, unfortunately, but, um, these vehicles have to agree on some kind of velocity in order to be able to pass through that intersection without actually colliding with each other So, consensus on velocity can be important for these kinds of tasks and others, and then there’s also consensus on location, which has a particular name, it’s called rendezvous So, consistence on location would be something like delivery drones that have to somehow agree on where the client is and be able to actually find the client to actually perform the delivery, or where the packages are so that they know where to route themselves to to actually pick up these packages. So, there are many different kinds of attacks that can be instantiated on these multi-robot systems I’m going to talk about one in particular that’s called the civil attack, and it’s an attack that’s easy to implement actually and hard to guard against, and that’s why I chose that one, and so, here, conceptually, these are ground clients, so you can think of them as people that have particular requests for delivery, etc., and your task is essentially, or our task is to deploy robots to actually provide that service to the clients So, I’ll call those aerial robots, and you can think of them as having some kind of a support structure for the clients It could be delivery, it could be communication coverage, it could be anything, and the idea is that even a single agent in that network, which Eduardo also mentioned this morning, even a single agent could be an adversary, and that agent could spoof a cluster of other clients in the network that may or may not actually exist, but the point of spoofing a whole node of other clients in the network is that they can influence the behavior of the entire robot network, and arbitrarily moving them, for example, in the environment to disconnect the network, to break the network, or to just reposition some of your robots in a way that’s beneficial to the adversary, and this is called, um, a sibyl attack, and the red nodes there, I’m going to refer to them as spoofed nodes. So, this is what it would look like in an actual implementation, so these are just some robotic clients here, and we are controlling this flying router or this flying robot in the middle, and the idea is that flying robot is supposed to service the benign clients on the left-hand side, but there’s a malicious client here, and we just pose this spoofing attack, where the malicious client actually spoofs a cluster of nodes So, there, the malicious client spoofs a cluster of nodes and, indeed, pulls our robot all the way to the upper right-hand side of the environment by producing these sporadic requests. So, this is what it would look like in a physical implementation Now, I want to explain what’s actually happening understand the hoodism so, basically, on the left-hand side, I’m going to talk about what consensus is on a technical level, and on the right-hand side, I’m going to use an example to explain it intuitively for, say, the traffic example. So, imagine that we have nine agents, and they’re all agreeing on something. In this case, they’re going to agree on their heading direction. This is what it looks like graphically Intuitively, I just zoomed in on that intersection problem, where you have, um, autonomous vehicles, say, that are trying to drive as fast as they can along the highway Now, clearly, the optimal is for them all to have this same

heading, because if they don’t have the same heading, they’ll crash crash. Now, if there’s a sibyl attack, where now, we’ve injected some spoofed agents that have changed their heading here in red to coerce the rest of the team to do the same, what it would look like graphically is if the green are the heading angles of all the agents, and the spoof’s node has a heading angle shown in red, then over time, what you will find is that all the agents will try to align their heading as close as possible to the nearest spoofed agent, which can clearly create problems for your, um, driving task. So, this here, this gap is the difference between what the agents should be doing and what the spoofed agent is coercing them to do So, that is precisely the influence of having the sibyl attack on your network, and this is something that I, in my research, try to characterize and understand and try to find ways to minimize that as much as possible. So, how do we actually do that? Well, I mentioned before that with these physical networks, we have a new piece of information that we would be able to use, meaning the actual communication messages themselves, and, so, my idea is how can we take that information and actually make our objective, um, or meet our objective of bounding the influence that a sibyl attack could have on a multi-robot system? So, here’s the intuition, um, in terms of how we actually do that and what the approach is. So, imagine you have any two robots that are in a network, and they’re communicating with each other Well, it turns out that as they communicate, these are physical signals, electromagnetic waves that travel from one agent to another, and as they travel, they traverse the environment, and these arrows that you see are different paths that that signal takes to get from point A to point B Now, if we had a way to measure those paths, well, it would look something like this. This would be signal strength on the Y and angle on the X, and there would be two large paths that correspond with the two paths that you see there, physically that have traveled the environment, and if we have another pair of communicating agents and we measured the same quantity, this directional signal strength, well, it would look different for that pair because of the environment and their relative locations with respect to each other. So, the intuition here is that you can think of this almost as a signature of the two communicating agents, and, so, what’s really nice about that is that while an adversary can spoof the content of a message, they can’t spoof the physics, the way that the signal’s actually traversing the environment, and so that’s the key, and if we can kind of dig into that and lift that up into our multi-robot systems and our controllers, well, that’s the objective, and, so, if an adversary populates the network with fake identities, and you focus only on the data, then it’s easy to spoof, but if you use the physics of the wireless signal, then, all of a sudden, it becomes a lot harder to spoof So, that’s the idea, but how do we actually get that? How do we actually measure this? Well, one way is to put some expensive hardware on your robots, for example, some kind of a directional signal antenna, but this kind of defeats the purpose, right? Because some of these robots are going to be small, agile drones, so ideally, we could do this with very basic equipment, very basic hardware that a lot of robots already have today So t, it turns out that we were able to do this, this is something that we developed here at MIT, and the idea is to actually create a virtual sensor, and what’s really nice about this is that we can perform synthetic aperture radar essentially on a small robot with using just a signal omni-directional antenna, which we see there in the photo, and, so, the idea is that as that robot moves through the environment, it takes snapshots of the wireless signal, and as it takes many snapshots, at the end, what we do is stitch all those snapshots together, um, using signal processing, and we emulate a

directional antenna along the motion of our robot, and so what we get at the end of the day is precisely these directional signal profiles, where you can sense directly, um, the direction of signal strength for any pair of communicating agents, and, so, if, now, you go back to the case where you have a sibyl attack, well, all the messages that are coming from that agents agent are now going to have very similar physical profiles to within noise, and so that’s what we want to capture, and now, we have a way to actually do that on very basic platforms. There are a few important properties of these signal profiles. One of them is that there is some power that’s going to come from the direct line path, so you can start to think about validation of reported positions, for example, and the other is that multi-path peeklers, which are the peaks that are created by structure in the environment, these are unique for different, um, for different pairs of senders or clients So, my work develops the theoretical and experimental framework for actually using that extra information that we get out of the signals and actually quantifying what that means for our coordination task, because at the end of the day, that’s our goal. So, going back to this problem of, say, consensus, my work focuses on how to define what the maximum influence is of a spoofer, and how to actually get that from the wireless signals themselves, and, so, one of the results that we were able to get, and I just, I’ll flash it up here, I don’t expect anyone to understand it, but I’m going to tell you what it means graphically, so what it means is that we were able to show that if we have transmissions from two different agents and we compare them, we can compute a quantity called alpha for each observation, for each signal that we receive, and that alpha has a particular property that it’ll be near one in expectation for a legitimate transition, and for a spoofed tran spoofed transition, it’ll be near zero in expectation, and, so, we were actually able to prove this from the signal properties, but what’s important is that it wasn’t just in theory that we were able to show that this holds, we were able to show this in an actual experiment So, as you see, spoofed transmissions transitions are very close to zero, and legitimate transitions were very close to one, and, so, this is a pretty, um, this is a pretty interesting idea, because now, we have a way to actually say something about a sender that’s sending us information just by analyzing the communication that is already existing in the network. Okay, so, what about the impact on consensus? Because while that’s nice, at the end of the day, our objective is to say, okay, how does this affect my bottom line, where my bottom line is performance of my entire network? So, in order to really, um, explain this, I have to tell you guys a little bit about the mathematics behind consensus, and so I’ll delve into this just very slightly, but here’s, basically, an algorithm, or a protocol, consensus protocol, and the objective is that all agents agree on their value’s X, and the W that you see there is the weight, and that’s very important, because the weight tells you how much I’m going to weigh the value of one of my neighbors in the network. So, here’s a plot of what that weight looks like If we look at our network and we think there might be some adversaries in the network, say in this case, there’s two, what we would want ideally is that the weight of all the transmissions coming from those adversaries, or the spoofed agents, goes to zero over time, so I almost ignore them over time, and that for everyone else, it goes eventually to the average value, one over the number of legitmous nodes, and for every transition transition, remember, now, I have an alpha alpha, and the objective here is to characterize if I have many observations, say the weight of my first observation is green, um, the weight of a spoof node might be red, if I have many observations over time and I think of, I keep a track record, so beta is my track record of all of my observations, then what I’d like to see is that over time, those weights start to converge to what I would expect in terms of we start to ignore the spoofed node messages, and we start to weigh

more highly the messages from legitimate nodes, and, so, we were actually able to show that, that with time, we can bound the probability that we make a mistake, we can bound the probability that we classify a spoofed node as a legitimate node and vice versa, and that was done by a particular construction, careful construction of those actual weight values based on the transmissions, and, so, what does that mean for consensus? Well, what that means is that we were able to characterize exactly what this bound is, and we’re actually able to compute that for a particular network, say with a certain number of agents, with a certain signal to noise ratio, etc , and this is what it looks like in simulation. So, if you look, remember, this is the same graph that I had before of the heading angle, and what you see is here’s the input of the spoofed node, so he’s trying to direct all the cars to a particular heading angle, this is the true average that the agent should be converging to, and what you see over time is that the nine different agents do, indeed, converge with the true average, or at least very close to the true average, because they can now get extra information from the wireless signals So, it turns out that this result is, um, is very new Actually, you guys are hearing it before I present it, I’m going to be presenting it in two weeks at CDC, so this is the newest stuff that we’ve done in consensus, and these are what those signal profiles actually look like, so these are real transmissions, these two are coming from the same physical agent, these two are coming from two distinct agents, so you can get an appreciation for the kind of information that we can now get over wireless signals, and this is just some other work that we’ve done, also using the same concepts, but for different multi-robot tasks, such as coverage and drone delivery, and this was published in the, um, international, um, sorry, in autonomous robots, and it was actually featured on MIT news last year. Okay, so just a few quick acknowledgments These are, um, some of my main collaborators here at MIT, also my lab over at ASU, so I wanted to definitely acknowledge them and kind of leave with a summary statement, and the idea is that security and sensing for these physical networks of multi-robot systems that are going to be coming up in the near future, they definitely present new challenges, that we have to take very seriously, and we have to start thinking about what those vulnerabilities are and how can we rethink security for these systems, and the other point that I want to make is that this work shows that the very things that make securing those networks difficult, like their mobility, um, and their decentralized nature can actually be exploited to provide new powerful tools for security. So, in the future, I think it’s important to keep looking at what those vulnerabilities actually are for these systems that are on the horizon, that are literally arriving already, and characterize the information that can be obtained over these physical channels, because that’s something new for these networks that we didn’t quite have before. So here, I talked about physical networks of communication, but there are other kinds of physical networks that we can think of too, like, for example, vision, what’s more information that I can get from observing an agent versus just communicating over the Internet to an agent? What new opportunities does that present for security? And then, importantly, not just characterizing the information that I can get, but also understanding how to best use it to actually improve or secure my algorithms that these robots are going to be running. So, in summary, I think it’s very important to think about these new problems of robotics and communication and how mobility can help in terms of getting more information, and then also how that additional information that we get over the physical channels can help to improve our feedback, to improve the actual performance of our algorithms as a whole I’ll leave it there, and I’m happy to answer any questions (Applause ) >> SPEAKER: Any questions from the audience?

>> SPEAKER: I mean, honestly, the results that we had are really amazing and help us in many ways when building, like, blockchain applications, because this really tackles consensus for malicious agents. Do you think it would be possible to do this in a neighborhood basis? So, you would have, let’s say multiple agents, and they have different consensus based on the weights of the signals, so you would have a heterogeneous system, so do you think it’s easy to implement this considering the low power compute that they have onboard? >> SPEAKER: That’s a good question. Yeah, so, I think it’s definitely possible in terms of the computation that we’re doing here. It’s also pretty lightweight, because we’re looking at distributed systems, but I don’t know, I’d have to know more about the computational, um, essentially bounds that you’re talking about >> SPEAKER: Okay, I’ll just ask afterwards >> SPEAKER: Yeah, I think that would be nice >> SPEAKER: Any other questions? I have one myself So, for example, um, the results you have shown, like, basically, illustrate, um, these, like, problems with the physical signals, like in multi-robot systems, right, but for example, did you try, like, to replicate the same results with peer-to-peer networks? For example, we don’t have, like, a client approach, but more like a peer-to-peer approach >> SPEAKER: I see. Is there still mobility in that case? >> SPEAKER: Yes >> SPEAKER: Okay Yeah, so, for us, key is to have mobile agents that are actually physically communicating with each other I think if that’s still the case in the example you’re talking about, then, yes, it is compatible >> SPEAKER: Thank you very much. Let’s thank the speaker once more (Applause ) >> SPEAKER: So, I will introduce the last speaker of this session It’s Alexander Kapitonov He’s an assistant professor, and he’s going to talk about a fascinating project of, and I hope you enjoy his presentation >> SPEAKER: Thank you, Eduardo, for your words, but, um, small note, already associate professor >> SPEAKER: Congratulations (Applause.) >> SPEAKER: Thank you. Nice to see you here This is a really good symposium with whole different tiers in blockchain and applications are collected together, and I hope you’re enjoying the event, but my speech today will be about, um, the other side of the region region, and we call it robonomics. I will explain what I mean a little bit later Robonomics, the economy of problems In my opinion, um, there is an important theme to create the open market and the open market relations as much as possible, because this can help us to find the optimal position, optimal size and opt optimal way for the companies to create the products and deliver it to customers, and the same situation with robotics robotics, because it’s, like, it can be, like, a separate company, because it has, um, the solution center, it has actors and the sensors, and the product-like service or some delivery of foods and other things Okay, a little bit about me I’m working with robotics already, nine years and participate in the different international

events, robotic competitions, but for all of them, it’s, like, a red line , with all the systems right now, they’re controlled or connected, um, the the vendors, to the main vendors If we are talking about IOT things, there is a lot of solutions from Microsoft, Amazon They just released their Amazon mobile market, one of the services, and they try to connect all the things through the common gateway The distributed solutions, it’s one of the ways how to create the new relations and communications channeled between the robotic system, autonomous agents, and services , but look at this Um, for the people, the common interface for communication, it’s money. If I have money, I can interact with all the participants of the market Here is the first money for robots, in my sense, of course It’s only possible to work with banks, bank account, but if you, um, watch the movie, it was the situation when the robot come to bank and ask to open the new bank account, and they asked why, but right now, we know why, because they’re robots, they have find more autonomy right now to create a really long supply chain The second point, smart contracts, it’s a first formal description of the business process that’s suitable for robots The robot can read it and just can sign it, it depends from the situation, of course, but right now, the business process description makes possible to create the platforms like Air B & B and Uber, using just the codes as Eduardo showed A little bit about our history, because when I told about money for robots, our first case, um, in that, um, application field was, um, the drone service , and it was drone filming and drone photographer task We are organizing the first demonstration in, um, in the end of 2015, where it coordinates in making photo and sending the results with a smart phone, and it’s possible Right now, it’s possible for a more complex system, not only for the drones, but as you can see, the drone applications, right now, it’s one of the simplest and obvious cases, when the

mobile robots make service and make a product for the customer, and after that, um, we start to search the new areas for the implementation, and it was machine to machine economic interaction in the global sense, because mobile robots, it’s a good case, but there is a lot of tasks to do For example, sensing about global warming, a case that Eduardo showed, um, ecological situation, because right now, we’re not so clear for us about, um, ecological situations and environment state around us. There is some estimation, but the real situation is not so clear The last year, what we start to develop, it’s industry for zero application, because you know right now, the examples where the lights out factors exist , lights out factor, it’s like a separate robot, and it can be an economical agent separately without any problems There is more spheres for, um, applications of those ideas, ideas of a robot economy Of course, right now, we want to know how the product is made, because, um, it can be really not so clear what’s, how employees develop the product and what sources are used for that , and technical supply chain is really needed right now The next application, it’s autonomous service provider As we, um, see, the previous speakers mentioned mentioned many about the drone logistic and autonomous logistics. It’s, um, a really complex task, but look at this, we have many different agents in the world, like separate countries, separate, um, companies, and they should communicate to each other, but it’s, right now, that’s impossible, that some, for example, um, one country collects all the data about the, um, maps and digital elevation model It should be local solution for each country, or maybe for each company which provides the flights This device, right now, it’s really possible, and I saw some implement implementations when, um, where, um, for example, you can add your sensor measurements about CO2 or dust pollution and publish that information in the public network, or sell this information to the big vendors. Also, it’s possible to organize the video monitoring, placing the cameras which are controlled by us Each participant can just place the camera and send the data to the main center. Of course, as I said, industry for

zero, it’s one of the ways where we, um, are thinking for it and developing the real solution Right now, um, a little bit deeper about technology and how we, um, are solving the problems of the communication and the standardization of the contracts, of the data In the basics of, um, robonomics, or economy of robots, um, it’s two ideas. There are two ideas The first one, the first new institutional economics, he was one of the first who proposed about, um, transactional costs, and he said that the open market tried to find the best size or best solution for the company , what relation should be inside the company, and what relations should be outside the company, because right now, you can see, the big companies, they have, um, many different transactionalactions inside, and sometimes, it’s much more cheaper to make a deal with a small separate company to solve the problem than to try to develop solution inside the big company There’s start-ups, which around the Google, there’s the same, and the second idea , owner of that idea, he’s proposing, um, the solution for the USSR economy and create the global system to control the USSR economy Maybe, it was helpful, but it’s not the bed point Combining together those ideas that robots, it’s like a separate company, and the open market for the robots right now is possible, we’re trying to deliver open market relations to the autonomous systems systems, and, um, for the technological solutions, um, for those ideas, we are using, um, the operating system. Do you know the operating system? The operating system, right now, the fact that they’re standard, not only for mobile robotics or, um, like, separate sensors or systems, but also for industrial systems, I haven’t seen the solutions right now on the market The second one, it’s blockchain, as I told, we started to work with, um, blockchain and robots implementation three years ago, and if you know, blockchain was the first platform with a real workable smart contract s and with a good price, of course We had developed a particle for the operating system and for blockchain and for the additional communication channel, because, um, they call it published subscriber, it’s a light channel without logging the information inside the blockchain and inside to the main network

We’re still waiting for the good solution for whisper, because it’s really needed in the blockchain, and to standardize those things, we’re using , we start with the docket containers, but right now, decide to shift, because there is a lot of possibilities for the formal description of the software inside This is the, um, configuration file where you can describe all the things, and it’s, like, it can be like a passport of the robot That’s what we are using for it Um, a little bit more about market mechanics and all the things, what we are doing with blockchain and the robot operating system. The first point, it’s a cyber-physical system, robot operating system supported, and the users, here is the human, but, um, the user can be not only the human, but also another cyber-physical system, and we are supposing that they supply to their light channel communication Right now , after that, we have that, um, smart contracts for blockchain, which allows us, um, to find the similar endemic and create another special smart contracts to make a deal with the customer and cyber-physical system After that that, when the liability contract is signed, the cyber-physical starts working, and the result s , it’s observing network, it’s like, um, miners in the blockchain, but it’s a specific, um, systems For example, it can be the net network of the sensors or video cameras which can prove that your car, your autonomous car was here or not, and if observing networks sign the contract, it goes to approvement, and the observing network gets the award and the cyber-physical system getting the payment from the customer. This is the main principle of the work with the blockchain, and right now, I show you the pre-cases I hope it’s not so complicated and it will be obvious, how do we use the market mechanism for those things Of course, as I told, the supply chain for the drone flight, it’s one of the big problems right now, because you need to get information, you need to get the license, you need to register your drone It’s kind of not so easy, but it’s necessary So, those things, the blockchain is much more suitable than the systems already exist, and here’s the demonstration of the first drone launch, what I told you before A little bit about the situation,

here is the — music playing — it depends from my position or not Here is the consult, where our program makes a contract with the drone, and you can see, it’s, um, a real Russian winter, but our programmer looks like fishermen, but there is not. Maybe, he’s a fisherman in the blockchain, but not, like, a real fisherman, and here, we demonstrated how to, um, send the specific GPS mission, um, to the drone with a smart contract, and it was three years ago, and right now, we develop additional solutions, and I will show how we use it for the industry for zero and for the communication with robotic system The next example, it will be about industry for zero, and the idea, it’s related, of course, with lights out factories, and to make a much more closer connection between the production line and the production process with the investor, because right now, it’s not, um, there is no direct connection with those sites, but I think it’s really necessary, right now, we’re trying to organize start-ups and all the things , and we want to make the direct delivery of the thoughts to the production, and here is the demonstration, this is industry for zero stint, it’s a big technique station where we’re modeling the situation There is the two sides, one is source of materials, and the second, it’s, um, storaging for the products, and here is the interface for the investors, and you are like, investor? You invest in the special product, and production line tries to find the optimal position between the, um, demand from investors, because the different investors, they want different products and different value , and after they’re getting the , after they’re putting the investors on the market, the production line starts to work, and the material source delivering to the production line and to the storaging Okay, and the last demonstration, it’s a robotic art. Why? Because it’s one of the short ways to demonstrate how the robot, um, is making the final product, and, for example, the picture can be, like, the final product, but here is a lot of questions We made the such interface where, um, the robot will tell you what picture will be painted painted in the few minutes, and the investors or, like, customers can vote for the picture, but the picture is made, um, by analyzing, taking the hot topics for

the hash tags and putting it inside the, um, AI , and in the final way, we are getting the two words, and words translated to Chinese, in that case As you can see, there is several pictures on the background, and look at this, there is no human in that supply chain of that picture production process, and there is a really big question, um, how we will deal with it, because there is the owner of the picture, the owner of the of that in that case, and actually, you know the situation with open source, right now, there is a lot of open source license, and maybe, for the robots, we should create the same, like the open robot license, like the developers, um, given the freedom for the robot that they made.P I think it’s possible, and blockchain and the open market relations, it’s much more suitable for that situation, where the robots own it and try to improve the value of the global economy. Thank you for attention. I am ready to answer your questions (Applause ) >> SPEAKER: Thank you. I think we have time for one quick question >> SPEAKER: Lovely presentation. I just wanted to ask one quick thing regarding the robot art, because in the blockchain space, there is a concept of artonomous Its are a robot that runs in a Dow but basically owns himself, and all the art he creates, he sells it to other people, which basically gives him money to either continue to create art or to invest in something else So, basically, a Dow seems to be very natural to be a way that a robot can own itself >> SPEAKER: I think it’s very similar, and there’s no problem to organize the cooperation with that project >> SPEAKER: But it’s, like, self-ownership of robots >> SPEAKER: Right now, there is not, um, like, a legal paper that exists It just exists, but there is a lot of legal questions inside that If you’re checking the design for robotics, it’s, um, the big book from the IAAA society with description of all the problems, social problems, etiquette problems of the robotics. There is a lot of questions for the robot art For example — how we should deal with it Maybe in team robot, we’ll be much more, um, the first robot who will own it by itself >> SPEAKER: Okay. Let’s thank the speaker once more (Applause.) >> SPEAKER: Let’s go for lunch Lunch is ready in the other room, so let’s get back here at 1:00 p.m , because we have a lot of paper presentations. So, yeah, see you in a bit >> SPEAKER: Who’s also — we work together on this project to

get these robots, which I’ll now pass around, we

call them the Billy robots, to get a glob to fit on this

system. This is an ongoing project — off mic

great. So, the robot moves by locking its back foot on to the

lattice structure and extending its front foot in front of it,

and then it follows with a back foot. Since it’s locked to the

structure, it has very few motions in which it can take

It can take a step forward, a step forward and right, a step

forward and left, or a step two in front of it. So, using those

sort of primitives, Amira showed a simulation of how this robot system might be able to build these lattice structures, and so in this case, the robots are returning to the red material depost, where they’ll each collect one and then extend them into a structure. So, that’s a longer-reaching goal of the project, and our goal where we implemented the blockchain was just a mapping and exploration section. This robot that we used has a long history It was initially introduced in the early 2000s, and then it was adopted for these lattice structures by our group, and you can see the actuators here, there’s five to make it look amote, and then one for each foot to lock, and then there’s only two sensors on a block, so it can’t actually detect a movement unless its foot is in place, so it has very limited environmental awareness. This figure will show the locomotion of a robot, where it steps forward, steps forward right, and you can see there’s one of these voxles is missing, so until it will put its foot there, it will not trigger the switch, but it knows its foot should be there, and that will be marked as a void voxel In our implementation, we fill in a matrix with zeros and ones to show if the voxel is present or not present. The hardware we developed to run the blockchain is pictured here, and it’s a pretty simple little embedded system, just a chip, a driver, and a radio for the robots to communicate with This was expanded from the — as we were thinking about how can we make this system use blockchain and be useful, we thought about, um, two things The first was that we intend this for space application. So, blockchain can be really useful, um, because it’s distributed and robust, and space applications, your rovers can go everywhere, you don’t know what’s necessarily going to happen, so having the entirety stored on each node can be very useful when you’re trying to pull the system back. You can go to whatever robot is still functioning and is still, um, easily accessible, and you can bring it back and get the entire data and the history of all the robots. In our blockchain, we wanted to include not only, um, the, like, presence of the voxels and the map, but also some, like, data about the hardware and how it performed, because that’s, um, definitely of interest when developing, um, rovers, and that led us into this concept we were calling proof of validity, and so that’s, um, sort of tailoring the blockchain to make it useful to the robotic system, as well as, like, the data that you’re trying to store. So, one of the things we were thinking about was, um, in scientific data collecting, you always take multiple measurements, and until you have some statistical significance, you don’t claim that those measurements were valid. So, we’re using that to do the mining in our system, so by taking several readings, um, we can then mine the blocks So, in our implementation, each voxel maps to one transaction, those transactions are stored in all of the robots, and when we have two redundant transactions, um, with the same reading and different robot IDs, you can see how this could be expanded to measure other things that might get greater variance, um, than, like, a bullion. For instance, like, temperature and actual typology of a surface So, implementing that algorithm Um, Amira did another simulation to show the robots actually mining these voxel structures So, the blue have been mined, the red have been explored by one robot, and the white voxels have not been explored by any robots, so the simulation, um, and the robots will run until the entirety of the voxel structure is enclosed by a ring of voids or zeros in the matrix, and you also see that in the inner sections, where there are no voxels, those get filled in as well, so you’ll know the robot will never be able to explore those areas. Now on to John >> SPEAKER: Thanks I’m going to talk a little bit about the implementation, specifically for the Billy robot, and I call this the trade-off between relevance and faithful representation. I’ll sort of explain what that means So, to our knowledge, um, there are some existing blockchain things related towards embedded platforms, but the thing that’s sort of unique about our system is each Billy

is sort of a node in a network, so to our knowledge, there’s no existing implementation that’s immediately compatible with this scheme So, one method you could adopt for implementing, um, something like this is you could look at a popular existing version of blockchain and try to port it over bit by bit and just match it exactly This is, like, one, it’s very time-consuming, and because the application is usually not through robotics, you’re going to be spending a lot of time reporting over features that aren’t necessarily useful for your system, and the other major drawback is a lot of these existing are usually made for PCs, or, like, things with more, um, hardware spec, so just sort of to remind you, she already talked about this, but these are sort of the, I think of it sort of as two categories, there’s sort of a limitation and processor speed and storage, and also one in communication bandwidth and bit rate, so for our particular system, the, um, arm press is not particularly fast The thing to note about this actually is there’s no, um, non-voluntary storage for this particular, um, implementation, and that’s not really a good thing, because that means if your robot is paired off, you lose all your data, so that’s obviously a really big drawback, and the communication protocol we’re using is sort of this low-powered radio type module, which, you know, is very good for embedded systems, because you’re going to be mobile, you’re going to be running off of battery, but that comes at the expense of, um, sort of bit rate and bandwidth So, this equation sort of just helps me think about, if you’re just a traditional robot scientist and you only care about making a robot do robot functions, it basically describes you only have so many cycles that your processor can run, so if you spend a lot of it doing these blockchain actions, you spend less time doing what you traditionally think of as, um, robot functions. If you’re less than 10 percent, then you got to really think about, or less than some substantial amount, you really got to rethink your implementation of blockchain, and first, particularly, this is, like, using different hashing algorithms and changing the type of data we’re storing So, in terms of the memory storage, one way, just pictorially, how you might think about it, as it turns out for existing robots, a lot of these type of robots don’t actually use a lot of the memory that’s available In fact, for the Billy, it’s less than 20 percent, so that’s really good indication that you might use some of this extra memory for things like the blockchain, and however, this sort of is to, um, harking back to this idea of existing implementations, a lot of times, these blockchains can be over gig bites in size, and obviously, that doesn’t meet the constraints of the Billy, but one thing you can do to make this better is if you go and choose your blockchain data structures in sort of a clever way, which in our case is this lattice structure, you can kind of overlap this existing area, where if you’re just a traditional mapping robot, you would have already been storing this data, but now, you can pair the two, and you get more existing room to store blockchain information Obviously, with increased storage, you also need to increase your communication bandwidth, and this is sort of, one way you can think about it is if N is the number of blocks you’re expecting in your blockchain and your block size multiplied by that gives you some amount of memory and then there’s sort of this preamble that with any communication protocol, you need it in order for different nodes to transmit and receive, and so this can inform, like, once you do this calculation, what your packet size is, so that informs what kind of communication protocol you might choose. The other way this is really useful is if you are limited to certain communication protocol, you can go and find out exactly what the size of the blockchain size is it uses, and in this case, it sort of limits you towards a finite size for each So, yeah, some of the other communication protocols you might consider traditionally, WiFi, that’s pretty common for PCs. LARA is the one we’re using The Bluetooth looks promising, but there’s some issues. You really don’t want this action of pairing to go on, because if you want to send a new robot in space and want it to be added to the network, you can’t be pairing it remotely >> SPEAKER: So, given this and given the hardware that we developed, um, and we started trying to tune the blockchain, so what data are we trying to store in our blockchain, what data do we need, and what can we send, we found that the entire, like, the packet size is the length of the first stanza of Lucy in the sky with diamonds We tried to say can we make a blockchain small enough to fit into a hundred bytes? The answer was no. We found we could include four transactions So, if your voxel structure was four, it would be enough Otherwise, you need do a multi-packet communication protocol, which might be,

you know, obvious, but, um, I think this sort of, um, this concept of tuning your blockchain for your applications is one that hasn’t been particularly well-explored. Um, so now, we’re looking at future systems, and given that we developed a blockchain to fit on this robotic system, if we just changed the hardware a little bit, can we develop robotic systems that will fit, um, co-fit with the blockchain? So, thank you (Applause.) >> SPEAKER: I think we have time for one quick question >> SPEAKER: My technical knowledge really, um, kind of stops me from understanding the sophistication of it, so I’m just going to try to ask a very simple question, which is are there any other systems than blockchain that you could have used to perform the same functions? And if so, why didn’t you choose those? I’m just trying to figure out this distinction of where to draw the line between where blockchain is actually an interesting case study >> SPEAKER: Absolutely. Yeah, so, there’s a bunch of ways to decentralize robotic systems, and we could have used any of them much more easily, but I think what we were really interested in was how to push the blockchain technology into the robotics field, so that’s kind of what we aimed to do with this project We can talk about it more, if you’re interested >> SPEAKER: Thank you very much (Applause ) >> SPEAKER: My talk today is about using blockchain to make a better social network. Sharing is a fundamental human experience, and we’ve seen a lot of that with social media. What has traditionally served as a way in which we can connect with friends and family via offline conversations, now social media has kind of amplified that Some statistics from this year, that 30 percent of all time we spend online is on social networks, and teenagers spend up to nine hours every day, um, on social media, and so the question we’re asking is is this necessary and is this important, is this helpful, and there’s a lot of literature about how it’s changing the fabric of human society, and this is not a symposium to discuss that, but there’s also some work done in the psychology of live stories, and that lens views social media as a form in which we disseminate live stories, so regardless of how we personally feel about this topic, social media is here to stay, and in this paper, we’re talking about two main things The content that we see on social media, surely, I know what I have posted or what I have read, but if I come across something that has maybe gone viral, it’s hard for me to track down where that came from. Sometimes, there is information about it, sometimes, there is not Also related to that, there is a question of whether that person intended for their work to achieve the skill and the scope that it has reached by going viral, and so is there a way to control that? Is there a way to ask who first wrote, created, posted this content? And if there is a way in which we could monetize this, because in the arts, there have been certain

gatekeepers of artistic intent If we want to, maybe, submit a book for publication, people have tried to subvert that by maybe posting something on YouTube, and that has proved to be successful for them. What we are trying to ask is for ordinary users, like you and I, is there a way for me to make social networking, um, financially viable, is there a way in which I can be rewarded for my content, and also a way for me to control the region scale of my content The fundamental question that this poses then is digital identity. How do we determine identity? If we think back to what we have in place right now, we have a user username, maybe a profile picture, and an e-mail, address, or phone number for authentication purposes, but we have to realize that I could create multiple accounts with multiple e-mail addresses as the source account of those, and, so, once again, the question is how much of my anonymity am I willing to disclose? So, we are in a spectrum of anonymity, and on one hand, we have profile-centric networks, like Facebook or Linked In, where everything is tied to a profile, and we have non-anonymous profile-centric networks, like Facebook and Linked Lin, and anonymous networks, like Whisper, where there’s no way for us to ascertain who posted something and who is this person in real life, and, so, many social networks are on this spectrum, somewhat in the middle, where, um, some require us to post some information about ourselves, others do thought, not, so that’s the first aspect. The second is how much is the user willing to disclose? Maybe for my profile picture on Facebook, I post a picture of a blue sky day and using a fake name, and nobody would know So, on this spectrum of anonymity, is there a way that we can bring in blockchain or the framework of blockchain-like systems to establish digital identity, to control the scope of my content and to monetize that concontent? So, this is the framework for determining digital identity. As I said, there is the, um, e-mail address, phone number, profile picture, a username, and a network identifier, like a hash tag or the address mechanism. What we’re proposing is a blockchain-enhanced version, BEV, for social networking sites The competent that we’ll be using, AIML algorithms to tune my sharing preferences. So, maybe it is that I don’t want to share as much with everybody in my network and limit it to a subset of people. Can there be some algorithms that detect that automatically? And since this system can potentially be used for monetizing content, I stand to gain reward, so are there algorithms that create rewards for me? And allow the users to share preferences. Here is the general framework The user is interacting with different kinds of social networking systems, and these social networking systems are composed of other users, system APIs, which are always gathering data, like the keynote early this morning, where, um, Sandy talked about the ads that are tailored based on the content that a user has accessed or created on other websites, and all of this is stored in a data repository, it is stored in the blockchain further, and there are vetting algorithms to see if that content is appropriate, because there are some websites with explicit rules for how you have to behave on that network, so vetting algorithms. There are also algorithms for preferences, how much do I choose to share with which subsets of people If there are systemic APIs that are using my information without my consent, which is currently the case for most networks, um, there’s a way for the blockchain enhanced version to track such queries, and then

there’s the rewards archive that ties in, so maybe I share this much, and I get this much of a, um, incentive, again, in the form of bit coin or somethinginate to have that framework More of this is explained here here. So, we have two distinct systems, one is the user-generated data and how it interacts with the blockchain levels on the right-hand side So, when a user generates data, the data is raw data, and then all of that is stored in the blockchain and its repository. The amount of that raw data that is accessed by other systems, as someone shares it, someone re-tweets it, um, or it is going viral somehow, so all of that is stored in the shared data It also shares, it also stores how much I share of other people’s data, so it’s a share data archive, and all of this can be attributed distinctly, because blockchain has ways in which, without dispute, I can attribute content to insert a note. In the second level, I can tailor my sharing algorithms, so this is where some, I can use some AIML algorithms to vet that content, and I can say that I’m not going to share content coming from certain websites, so that’s the rule I said, and then the node automatically applies it to, um, content from that website or that friend or that node and similar other nodes The third level, there is the shared data archive, which is also using a lot of, um, machine learning algorithms to detect how much of, um, how much of the data is being engaged with by other users, so what kind of content is currently popular that I post In a way, I can get a sense of what content I post to get a more number of likes or shares or re-tweets, or is my content mostly sitting flat, and nobody engages with it, and the last one is the reward archive, which keeps track of the number of rewards I’ve gotten over time So, if I tend to post, let’s say jokes about, um, a certain kind of cookie, and if those jokes are more widely accepted than others, then it could be that, um, my rewards archive stores that information, and that is, um, communicated with the sharing archive, so over time, it can infer for me what kind of content gets what kind of response from other users in the network. There are multiple applications for this All that I’ve been talking about, um, is the application of conventional social networking systems, where humans are on the network, and they’re communicating with each other As Eduardo pointed out earlier, Facebook or these networks do not generate any content themselves, it’s mostly the content that they generate is, again, tailored from, um, what users post, so it is applicable for conventional social networking sites, but also, we can envision something about humans and bots, like, um, he pointed out earlier about, um, a robot, a drone having its own Facebook page, and we said that it could be that a, um, a Rumba, a vacuum robot is posting its services on Facebook, but a human and a robot are collaborating somehow Humans, bots, and IOT devices, so robots driving self-driving cars and going to grab coffee with a human and a friend is not that far-fetched, and there are lots of soft robots in existence right now for, let’s say grief counseling online or answering help questions for technical issues, so they’re already in place, and if there is a way for them to become more embedded in human life by having their own profiles online and these bots, they post their own content, the blockchain enhanced version of social networking sites can be, um, extended easily to incorporate all of these different kinds of entities, because each of those will be required to, again, create their own node with authentication schemes, and therefore, even if I want to create multiple accounts for myself, all of that information is stored somewhere in a node, and in case, um, there is a way for those nodes to become non-anonymous, I can still trace the identity of those nods, and maybe if the

legal framework is in place, trace the identity of the owners Um, this is one possible communication system. My prior work is in wireless networks, so for those who are familiar with that model, wireless sensor networks are networks of small nodes that fit about this mid size, and they have a computed transceiver and a battery onboard, and the goal is really, um, for these tiny nodes to be scattered in hostile areas or places where humans are not expected to work efficiently, like measuring the plankton distribution on the surface of an ocean, and they’re expected to communicate with each other, and, um, to make decisions based on the data that they’re continuously collecting, so there’s an obvious trade-off here, because they’re very small in size, the battery is small, so you run a real risk of, um, doing too much computation and the battery dies, and now, you have no data being collected from that coverage area, and so we can use a similar framework here, because we are naturally concerned about the battery, we do not have to worry about the trade-off between computation and, um, performance, and because the blockchain framework is already expensive in the electricity grid, so there are, um, the framework allows for more computationally efficient, um, algorithms to run for such nodes So, if we have a cluster kind of network adapted from the wireless sensor network area, where you have hubs or cluster heads that are responsible for their own group of nodes, so you can achieve some, an additional layer of communication without the network being fully connected So now, we would like to talk about the limitations of blockchain The blockchain itself is, as of right now, it’s used for, like, say financial transactions, or transactions of, um, limited volume, but when we talk about social media, I don’t remember off the top of my head the statistics, but the amount of data that is generated by users in terms of the number of tweets or posts or picture shares is way larger, so we have to think about how the blockchain can be scaled to accommodate that influx of data and how to store it with those timestamps and make it more efficient for social media kinds of transactions In social media, for example, this framework was developed to consider anything as a transaction transaction. My current work is in computational social science, so I’m looking at how people behave on social networks, we are looking at big data sets of, um, tweets, of posts, and using some algorithms to infer fundamental human constructs, why do people share, how do people behave when there is anonymity versus, um, there is non-anonymity So, it’s important that, um, we’re able to see how the blockchain can be scaled for so much, um, data The second thing is engineering emotions In the 60s, a paper was published by Barbara Meeker, and she said that, um, everything we do in human society can be viewed as a set of transactions. If I’m being really nice to some person, um, there is something going on there. It might be that I expect a favor in return, not necessarily from the same person, but from maybe someone else, or, um, maybe I just get an award by being nice to someone, and that reward is intangible, and I can attribute it to my altruistic self So, based on a gain theoretic construct of attributeing payoffs, where I do something because I get something from the utility of my actions, she proposed a framework of six principles about how humans behave in society So, she said rationality, altruism, group gain, these are the six fundamental competents of the gain theoretic structure, and, so, all humans behave according to these, so they try to be their rational self, where they maximize their own gain, or they try to be their altruistic self and maximize somebody else’s gain, and I try to maximize a group gain, or I try to be reciprocal So, um, if the AIML algorithms and the

blockchain can detect that , um, it can enter emotions without me, without my involvement involvement. A question raised earlier was where does human involvement in this end for it to be called really autonomous, and these are questions that are related to the limitations of blockchain So, if I can engineer emotions, then there are two outcomes On the one hand, I can say that if I post content that is typically offensive, I’m not going to get any rewards from the blockchain Given that risk, do I still go ahead and most this content and say I don’t care much about the rewards, I’ll speak my mind anyway? In that case, the use of blockchain is voided, because I’m doing all this computation, and I don’t need the blockchain incentives to apply to my social media activity. On the other hand, I might use a network to engineer emotions for me, so if I have content that maybe creates, um, posts that gather a large number of likes or tweets and I’m getting rewards for it without me actually posting the content, this could be done, especially as we expect robots and such independent systems to fit right in with our society Um, related to that, malicious bots, it might be that a certain robot creates art, and if I know that it gathers a huge number of likes and tweets, I post that kind of content, so generating data for rewards, and rewards for bots, now this goes back to the idea of co-located spaces, so not just humans have social media profiles, a bot has a social media profile, whether it’s a, um, software bot or an assistive bot. So, if the reward is in the form of bit coin, or even if it’s translated to currency, does a bit coin have the same appeal for a bot as it does for a human? What is a bot going to do with that bit coin? Is it going to enrich its existence, its being in some form by probably creating, um, more computational capacity? What are the imp implications of, um, fraud? How do you punish a robot for becoming malicious? These limitations are related to the general sphere of data. There is some more coming out of it, especially after Jack Bulkin’s lecture where he proposed three rules for an algorithmic robotic society, so there is some more coming out of it, but, um, there’s more to be investigated, and, um, so, in summary, the current model of social networking does not allow us to control the reach of our content uniformally without us going into privacy settings and, um, explicitly setting those boundaries. It also does not offer much in terms of security and trust, and definitely, there’s no monetization. If I write a particularly touching poem about, um, some issue that has moved me, the most I can expect is that it gets shared multiple times, and I have my, um, moment in the spotlight for a little bit, but there’s no incentive other than that, but using blockchain, the user can, um, be sure that the content here that the robot creates is attributeed to a node, and there’s a timestamp regarding where data has come from, so the new research areas, this paper tries to explore that of computational social science and blockchain, and our ongoing work is that of, um, how people, again, my background for this work comes from the lens of computational social science So, if I like and share some content, does it mean that I’m motivated just by the content or by the person who posted that content? So we’re looking at models to do this, and in general, we’re looking at the nature and design of digital emotions and how they’re expressed in social networks Um, for example, a recent work I did on an anonymous forum is that of a website called Scary Mommy

I don’t know if you’re familiar with that, but it’s a place where readers can post their thoughts about motherhood and parenting in general in 255 characters or less. It’s anonymous, there are only three ways in which a user can interact with the, they call it confessions, is using three buttons, like, hug, or me too So, if I went and posted that my dog died today, I’d expect that I’ll get a huge number of hugs instead of likes, so using this, if we are to find out, um, how do people behave online, are people more, um, willing to accept hugs from strangers online versus, um, strangers offline, so all these questions and computational social science that are being studied by many groups, and using blockchain, I think there is a way to, um, establish identity better, and so that’s it, and I’ll open it up to any questions questions (Applause ) >> SPEAKER: Thank you very much. While the next speaker, like, gets ready, um, maybe we can accept, like, one quick question. If not, I have one myself So, do you envision a situation in which we create, um, an environment in which we, in which lying costs money, yeah? So, for example, like, all these, like, bots that are, like, re-tweeting, like, um, this fake news, right, so every time that you make a transaction in this, like, system that you propose, you are putting, like, some money, like, or some escrow, and if, at the end, we figure out that, like, the source of that, like a piece of information was not very good, you will lose that money? So actually, lying costs you money? Do you think that will descent deincentivize? (Off mic ) >> SPEAKER: Thank you very much (Applause ) >> SPEAKER: Good afternoon I’m Aspen. I’m from Germany I work there. We are primarily a logistics institute, and we are doing a lot of material handling other kind of scenarios, and we work primarily with industry 4 0 and cyber-physical systems and a lot of other related things, and my project that’s funding me to do this kind of research is providing information by resource constraint data analysis. So, this is a kind of communication project that I started, and I was forced to use blockchain, and that’s why I don’t have blockchain on the title of the paper, and it would be great if you could prove me wrong, and then I could plug it into my framework and check what other, um, technology that I could use instead of blockchain So, if you, if I convince you, then it’s probably a very good framework to use blockchain in this scenario So, to set the premise of my work, um, and to comment about that picture, that’s the only dance floor that I’m comfortable in, and it’s our research facility with the nice, um, industrial floor, and we have, like, um, robots that can drive around and pick some work stations and all these different things, and this is, like, a very standard, um, indoor material handling facility, and if you see those boxes there, I don’t know, um, so these black things are electronic boards that are, um, photoboard PCPs that can communicate and interact with people, so you can basically

send messages and make inventory on those boxes, so this is, like, the premise, and I want to see how I can communicate with all these different things inside and to try to, like, see what comes in the future. So, what comes in the future is, um, that comes in the future, so there’s going to be very close, um, communication/interaction between humans, robots, and drones that are going to irk work, and this is a research facility where I am trying to tell you about what is the premise that I’m working on So, what will we try to explain is social network things or machines or industry. So here, in this case, I’m trying to take social network industry, where I have a supply chain, and I try to have, like, asset-centric transactions happening in a blockchain, and that’s what I try to create here. So, why did I choose context broker? Um, if you heard about, um, smart cities and communication systems, smart contracts, this is more like a precursor of things where they tried to, um, send and receive information with context mediation, so you have an asset that is, let’s say a street light, and who are all having access to the street light, and who can publish information for this street light would be the kind of, um, software that runs, um, smart city, and, so, this is, like, a European-funded project, where they tried to bid a fully-blown context broker, which is a centralized context broker that can, um, handle a lot of, um, requests and other kinds of stuff, and, um, this is very useful in developing context of our applications, and material handling is also, our production facilities are also, um, very context-aware, and when you make them decentralize, they become, um, well-optimized, and they can talk, talking between machines becomes very important So, um, this is how the context broker looks. So, if this blue box that is in the middle dies, which is a piece of software, then the whole, um, system comes down to a halt, and that’s where the centralized systems just break down, and then, um, I tried to use another system, and that’s what is running in our research facility, which is the MQTT context broker, which is not a context broker, it’s just a messaging system where you can send and receive messages, and it’s very lightweight, and what I can demonstrate later is that the stack that we tried to use, um, works Um, to set another use case for this, so, the, um, production of automotives would become very, um, decentralized and decoupled, so it would be a lot of robots, there will be a lot of autonomous machines and small fiber physical systems that communicate and try to organize industry, and this is from a different research project that’s called called smart face, and this project develops several decentralized process optimizations, so we have, like, different robots that bring, um, the piece that you’re working on on to different kinds of work stations, so if you have a manufacturing industry for cars, you have the doors that are being fitted, and if the next step in the process is going to be stop, then all the robots would drive down and try to optimize the situation that the factory doesn’t come to a halt just because one of the work stations has died. So, this kind of optimization happens, and for this, you need some kind of decentralized algorithms that should be running, and what I’m trying to do is to bid decentralized communication using blockchain. Um, so, the motivation, so I need a consistent data store where I can have multiple entities commute, trying to give and receive information, and then I have, like, I need a trust, or control issues, who runs the database, and then, um, do I need tamper-proof log of what happened in the previous thing, so this is basically what are the requirements for blockchain, and that is exactly what I also have in the context broker that I am trying to build, and then I also have all the other different things, like organizing the different nodes that are connected between each other, and then do I have to have a special, um, infrastructure where I have to log all the processes, or does the context broker do that inherently by itself. Then I talk about how can you use decentralized message processing and why centralized message processing is not good and all this kind of stuff. So, the only solution that I could come up with where I can have asset-centric communication is, um, using blockchains, and therefore, that’s how my, um, requirements looks. So, I have drones, I have robots, I have low-powered devices connected to a router, and they all have to communicate and organize and try to work

together in a socially networked industry, and, so, my stack would look something like this There’s a communication deface, there’s a memory pool, and there’s a consensus algorithm, and the message box. So, going down into the stack, I have, like, communication that is working, and then I have, like, um, mem pool, — the next time is to have consensus. I have to try to agree on this transaction with all the other nodes that are in the network, which would mean that the robot has to agree on another robot’s transaction, so they can collaborate on the working procedure, whatever the algorithm that is, um, underlying the process, they all have to communicate and agree. So, I only talk about what the payload is going to be consensed upon, so the next, um, there’s this consensus protocol that is working, I’m using tinder tindermeant, which is a blockchain protocol that does not need any kind of mining, so it’s controlled by a set of validators. Here, I started using one of the biggest, um, applications for tinderment that has been developed in the past, and then there were certain, um, things that I could not do, and we developed ABCIA application ourselves, sopeople can plug in their own decentralized applications in it to run, so therefore, it would be basically, like, a node that’s running on the robot that would connect to the ABCIA application and then start pushing messages or transactions So, that’s how the tinderment process works, and in my, um, tests, I use, like, a set of validators that are in a cloud server, so I can basically monitor a lot of, um, things and look for cost, how much cost to run such a blockchain, again again, to a centralized context broker, and then there’s this API and inventing that all happens through the interface, so you basically are running everything locally, and you’re trying to query different information within your local instance. So, what happens is that the data gets replicated throughout your network, so all robots agree using the consensus consensus, and therefore, you have the same data in all of the different nodes, and then you can just query upon this data as you’re querying would already be. So, one of the other problems for us was that we want to have to use mongo DB, and we also want to have to use, um, this communication message processing, so we tried to just use mongo DB and replicate all of our, um, data throughout the network, and in the end, what we have is that we did not change the whole stack, we just installed a new piece of software that would just retroactively work on the whole system. So then, this is, like, how the API and the inventing happens, and, so, you can plug in any kind of systems, let’s say a routing mechanism for the robots that is working outside of the robots, whether the routing , when the routing is not centralized, you can plug in a different one, or routing work stations or anything for that matter, and that’s how you would create a transaction At this point, we still have to store the data in the blockchain, but the blockchain is not public, so there, you would not have the problem, but then I have a few ideas of fusing IPFS or such similar systems where you can just put the URI of the data that you want to store, and then you can query that up. So, this is open, and that’s why we haven’t still open sourced our software, which is a very prototyped version, and we would like to have to want to open source it once we have the, um, let’s say an alpha version that really works and it’s ready for public use So, the overview of the decentralized context broker, the stack would look something like this, that you have all of these, um, pieces of software, the same piece of software that’s running with one file that has to be shared between each of these system, which is the genesis file, and then you can just, you have data replication basically, which you can then write, um, different decentralized applications. So, I tried to put all of these in these robots, and then we tried to run these systems We have, like, some kind of, um, let’s say performance statistics

that we have, and that’s how the, that’s how the stack would work, so once you start initializing the system, you would have, um, you would get all the data for joining into the blockchain, mostly, like, authentication data and the genesis file when the blockchain started, and, so, here, we would just send out peers that are, um, behind a domain, where they are load-balanced, so you basically can, um, have one validator, and you can basically discover all of the others once you join the main network, and then the system is ready for messages, and you can start sending messages in there, and what we did is we started sending 10,000, a hundred thousand messages on to this interface, and also compared it with MQT, because that’s what we use in the system. So, that’s what happened. So, every block has a thousand transactions that it can store and create consensus on, so what happened was that MQT performed way better when we sent a hundred messages, because there was a, um, time for commenting all the blocks in the blockchain, so it’s always better, but then once we started doing a thousand and 10,000 messages, this was consistent, and what happened, interesting thing that happened was that MQT stopped working once we started sending out a hundred thousand messages into the system, because it’s basically a very basic socket information system, where the socket gets overloaded, but what happened on our context broker was that it also broke down, but the problem, how we solved this decentralization also had, been trying to send the message out to a different service, and then this message would take on these messages. So, we could eventually get to this hundred thousand messages, where opposed to the other one, the whole system just broke down So, this is what we have, so we can still run a system that’s highly scalable, and this is just to show that all of the systems use the same amount of, um, usage, so there are, like, four servers on four different continents, so that’s, like, for the lowest peak there, I don’t know if I can use that, maybe not, so the lowest peak there is to show that, um, what happens when you send a hundred messages, and one of them is receiving all the messages as an initiator, and then, um, what happens when you send, like, um, a thousand and 10,000 We were not able to, so, the next large peaks that you can see there are for a hundred thousand, since it was not, like, a successful way of sending messages, we did not report that as valid test When you have a custom-made car, so if you have a custom-made car and you’re trying to order that on a web interface, all the parts would be assembled in different steps, so that is, like, a realtime view of how it happens in the industry, and then, um, there are, like, different steps that happen for the customization that you want, and this all happens in realtime for different parts that you want to put in your car, the safety things and everything, and then all of this, as and when the car is being built just for you l, all of the messages is being logged into the blockchain, and what happens later is you can check how my car was produced Of course, you’re not interested until there’s a breakdown, so you want to find out if the breakdown was because of you or because of the production, and then it would become much easier to figure out, um, during maintenance, and also to figure out all the different cars that had the same errors that can be used to check, um, check for this and to, let’s say avoid future failures in the production line, and you have a happy car at home. So, this is, like, the outlook of how things happened So, we have a decentralized system where we can send messages, we’re still looking at how to use the semantics and also how we can have adjacent messages to send into the system, if we want to do that or just not do that, that depends, and then, um, so that’s going back to how the stack works and how many messages that we can send and how easy it is, so the consensus part there can be replaced with anything else that does consensus, and it is important, like, what kind of consensus that we want to use, and currently, I use tinderment, and then, finally, I have this small demonstration where, um, I try to show you all of this working in different use case scenarios inside a research facility where this could be useful

That will be it (Applause ) >> SPEAKER: That was really fast, but if you have any questions, I’m happy to answer >> SPEAKER: It was super fast, but we’re running behind schedule, so maybe we’ll take the questions, like, offline, right? So, I’m going to ask the next speaker, like, to, like, um, go there >> SPEAKER: So, for those of you wondering who the horrible golf monsters who’s disturbing all the presentations, I’m Alex I’m going to be presenting Grex, decentralized hive mind Um, so, first thing I’m going to mention is I work for a company called Crya. We’re from the Netherlands, and basically, we’re a blockchain studio. We help our clients, um, shape the impact of decentralized technology. So, um, we help them from the ideation phase to the proof concept phase, and we take, um, mostly a practical approach, so whatever the concept asks for is actually what we’re trying to use, so we’re not really tied to any platform Um, we also have internal projects as well. The internal project we’re going to talk about today is Grex In February or April, I think, we participated in a hack-a-thon, and within 48 hours, we had to, um, yeah, make a proof of concept So, for today, we wrote a paper about that proof of concept We wrote the paper Grex, decentralized hive mind, and, yeah, we took some inspiration actually from a paper from Eduardo himself, the paper about, um, a new framework for decentralized swarm robotics, and our initial instinct was to just look around at the field swarm robotics and just see what’s going on over there So, um, first thing we saw is there’s a lot of applications for swarm robotics, applications such as, um, just exploration, maintenance, search and rescue, and despite all of these applications and all of these wonderful concepts and ideas, um, there is a lack of, um, actual applications in the real world. You don’t see a lot of swarm robotics applications when you walk around the city, so we were wondering, why is that? After looking at some literature, um, we found that, um, there are a lot of challenges that swarm robotics faces First thing, um, that made sense to us to identify as a challenge was, um, hybrid systems. So, you want a balance between centralized system and a decentralized system, you want to balance the drawbacks between them. The second challenge is actually transparency and trust So, this was mentioned in earlier presentations, and basically, what you want is you want to be able to trust the system, either mathematically or just, um, in a more organizational way Another challenge was

sustainability. Basically, when you have robots moving autonomously, you want them to not be able, to not not be able, um, to move around, you want them to be mobile, and you’re going to need batteries for that. Another challenge is physical implementation. So, a lot of the research we looked at cited physical implementation as a challenge, and we also saw that there is actually a lack of physical implementations of all these concepts and ideas ideas. Abstract intervention is also a challenge. Basically, when you have a decentralized system, it’s really hard to see what your system is actually doing Finally, scalability, which is just how many agents can it handle, so all of these challenges, out of all of these challenges, we were going to tackle these three, and before we tackle them, we actually looked at some similar research, so there were a few projects that were a little similar to ours, so we looked at what distinguished our project from theirs, and most of the research, most of the similar projects we saw, ethereium was the platform that was being used. The scalability, as a result, was tired to ethereium as a blockchain platform, which is around, for a public net, it’s around 15 transactions per second, and there was a focus on bozantine agents, so those were the things that were discussed in the relevant research So, we decided to put our focus on different things, so the physical agents and simple hardware weren’t really a point of emphasis in those, um, similar researches The ease of use, um, wasn’t discussed as well, and basically, the scalability, um, wasn’t a point of emphasis either, and there’s also the abstraction, which was one of the challenges I mentioned earlier So, the first thing, um, about our focus points is physical agents and simple hardware We decided to use simple hardware, so that’s a relative term, but we decided to go for single board computers, cheap drones, and just open source software As you can see, we have a construction over here, and basically, the top layer is tinderment big chain The middle layer is, um, ras berry pie, and the bottom layer is drone, and all of these things are basically just available for not too much money. So, we decided to use this for the experiment we’re going to show you in a second. The next point was we wanted the platform to be easy to use, so, um, the application uses ORM We’re using big chain as our blockchain platform, and they have an ORM driver, and basically, what ORM is, it’s an object relational mapping, and it’s a technique that lets you query and manipulate data from a database using an object-oriented paradigm, which basically just means you can approach it as an object, as an entity rather than just a set of transactions. So, I’m going to show you a little code. To use the ORM model, you first have to define, um, an agent model, so over here, we just have a name for the model, and we have some properties, and basically, all you have to do is fill in the properties you want to define for each entity Once Once you do that, you can instantiate an agent and say this is my agent, this is the data I want to use for it, and all you have to do is call the ORM driver and just instantiate your agent, and then, finally, once you have your agent instantiated, all you have to do is just put in the new updated data, and it updates it for you in the blockchain. So, you don’t have to deal with transactions or anything like that, you just approach it from a practical standpoint So, I have an agent over here, it’s just a physical agent, and I want it to also be on the blockchain, all you have to do is just, um, abstract what you want from it, what you want to know from it, and then just put it on the blockchain. So, the left point

of focus was abstract supervision, so to facilitate that, we built a front-end interface, which shows you where agents are on that. Basically, as you can see, the yellow dots are, um, simulated agents, and that’s just how it’s displayed in the front-end interface Finally, the experiment, so, the experiment we made with the focus points, um, we actually programmed that at the hack-a-thon within 48 hours, and the idea was to have, um, a random objective spawning, and basically, um, you respond random random agents somewhere on the map, and they start exploring. So, as you can see, those are the arrows on the image, and those agents start exploring, at some point, one of them detects something, and as soon as one of them detects something, all of the agents converge So, after running this experiment 420 times, I think, um, we came up with these results As you can see, um, the graph displays the max transactions versus the number of agents The throughput was lower than we expected, and we found out that was because of the bit chain DB ORM driver, and basically, what you get is you get a trade-off between abstraction and performance, so if you want to make it easy for people to use, then you have to trade some performance in place of that This next graph shows the detection versus the simulated agents There’s two, um, two types of agents, there’s the simulated agents and the actual agents, so we ran the experiment on a server, and we also run it on the single board computers As you can see, the means were pretty similar, and basically, what that means is the system isn’t really limited by, um, the actual hardware, so any limitations of the system, um, aren’t because of the single board computers, which is actually a good thing So, finally, what did we learn from this? Well, there’s a trade-off between performance and abstraction By using the bit chain DB ORM, the performance was a lot lower than the theoretical limit we had in mind, but, um, the abstraction we got from it was actually pretty nice, so you don’t have to think in the concept of transactions, but you just think about entities. Second, the bit chain DB throughput, um, was, um, lower than expected, it was around 20, 30 transactions per second, and it was theorized to be around 300 We also saw that low-power devices can actually run full blockchain nodes, and that’s great, and what we got from the experiment is basically just an open source, um, experiment platform So, in the future, what we want is, um, we want to create more experiments, and we want to make it accessible to everyone. So, you don’t have to be a seasoned coder to work with this bit chain DB ORM. We actually open sourced our code, and you can just use it to create your own experiments and do your own things with it So, the ORM we use provides a lot of abstraction. You can just use it to create your own experiments and just stay at conceptual level without having to delve deep into the code The interface we created allows you to monitor your experiments and see whether or not your experiment is actually doing what you expect it to do, and finally, we just want to use as many experiments as you can, as easily as possible. So, are there any questions? (Applause ) >> SPEAKER: So, I think that while the next speaker, like, prepares, because we, unfortunately, ate up the time for the coffee break, even though there’s fresh coffee down there, guys, so if you want, like, this is your moment, maybe we can have a very quick question question So, maybe I can ask you something. I see that you

prepared, so, the base of your research was, like, 48 hours of coding, right? >> SPEAKER: Yep >> SPEAKER: So, what do you think are the myths, like, that people, like, tend to have when they approach these robotics, like blockchain, that you dismiddify through basically, like, getting these experiments and running, like, the code and creating the platform so fast? Do you think it’s basically the fact that people think that it’s just complicated,t, that you need to have a lot of tech expertise? What are your thoughts on that? >> SPEAKER: I think the most important point we can take from it is that, um, it is actually possible, but a lot of work goes into it, so when I say we programmed it in 48 hours, we didn’t spend the entire 48 hours programming it, but it was pretty close, but, I mean, it’s possible, um, things don’t always go as you plan, but it’s definitely possible, it’s not as hard as it seems >> SPEAKER: Okay, thank you very much >> SPEAKER: You’re welcome (Applause ) >> SPEAKER: Good afternoon I’m going to do a short overview about blockchain integration with robotics and AI. I will do a short presentation so we can keep up with the schedule, and I will divide my presentation in two branches, the robotic applications and artificial applications As we all see in the news and in the papers that are out there every day, thousands of new approaches are being proposed every year, but they lack standards The robotic applications is still pioneer work, like Eduardo is one of those guys that are doing the pioneer work. Starting with robotic networks, there are a couple of papers out there that present some work, starting with the problem of robotic recognition to keep up with the logs in a safer way in factories, and the solution is a proprietary work chain with faster validation speeds to keep up with the robotics. As a follow-up to this work, there’s the work done from Miguel, he will present his work at a later hour, so I will not talk very much about this, but what they do is they use technology tools to show that what’s been done in the smart contracts — they use intelligence to per enhance the quality of those systems. As was shown this morning, they presented a paper that can be used to solve issues in swarm robotics, and they present a set of problems that needs to be overcome so these networks can be used industrially They also have a robot chain program that tackles the issues regarding privacy of data, and the solution is an interconnective robotics rung that can change behavior depending on the data they receive, and it can improve machine learning algorithms and propagate them to blockchain without compromising the privacy of the data. I also included another paper that’s from Chin, and they tackle the problem about validation speed, and they use artificial intelligence to do it They use a convolution network that classifies nodes as a super random node that can be used to validate transactions much faster than the normal consensus, so that robotic systems can be used There are thousand of those, and the first one, the singularity

net, I believe everyone has heard about them, they use their solution, and the problem they tackle is the A AGI They are building a marketplace so you can sell algorithms, and they implement an API that to solve a specific problem The second one, they tackle the problem of time-consuming tasks that developers must handle, and they use automation with artificial intelligence and robotics so that developers can use their platform for programming and building solutions solutions They build a marketplace that can be used to sell and propagateidate propagate data sites and algorithms in a trustworthy way They also tackle the problem about slow transaction speeds and specific languages and the inflexibility of the blockchain, and they propose the solution about narrow network that converts scripts to the blockchain, so you don’t need to know the actual language to build the smart contract, but you only need to know how to write simple screens. They improve the security of the blockchain by associating rules with the smart contracts, so the developers reduce the number of bugs, and they use artificial intelligence and different consensus to, um, as faster evaluation speeds and changes the inflexibility of the blockchain Nar Marketplace can be helpful for robotic systems, and these can be very good, because most of robotic systems and artificial intelligence algorithms use, um, narrow networks and other systems that takes a lot of time to train, and this mining process lets the miners, um, receive money to train algorithms for others So, you can train narrow networking in minutes instead of days. There’s a conclusion Blockchain and smart contracts can and will be used to improve the connectivity of different blockchains and lead to robotic networks The articles for integration of robotics, artificial intelligence, and blockchain will pass by the energy and time consumption of the systems, and achieving realtime, for data transmission. If some standards are missing, like what we saw in the morning, different blockchains cannot operate with another blockchain, so we must tackle these problems before we give up on solutions This field of work will grow once these standards are completed. Thank you (Applause ) >> SPEAKER: Any questions? Now we have plenty of time (Laughing ) >> SPEAKER: The real time problem popping up, like, following your analysis, what directions or solution directions do you see to tackle that problem? >> SPEAKER: The first direction is the change of consensus, of course, because many of the blockchains have slow consensus, and we see a lot of proposals to improve the consensus and how many transactions can be validated per se, like the fundamental vice that will allow us to reach realtime systems, but there are a lot of other problems, like the Internet’s latency, like, we need to send the blocks and the information to other networks, to other nodes, and if the connectivity is slow, we will not be able to achieve a realtime system, so this is another problem that needs to be tackled, as many problems, but consensus is, like, the first

one that has been tackled >> SPEAKER: Any other questions? So, I have one myself Do you envision, like, a market pool? Like a situation in which, like, the market needs, like, this technology, like, so, it brings, like, researchers like to think about, like, stuff that we are not currently thinking, or do you think that the future of these, like, synergies, like a technology push, the fact that at some point in time, some new consensus will arise, and then, like, this field will flourish? What do you think? >> SPEAKER: Okay, in the near future, probably not, but in mid-term, like, the work that has been done at my university and my laboratory is tackling the problem that factories have a lot of different manufacturers, there are robotic arms and other robots that are built by different manufacturers, so if you log those to a simple database, you can alter them, if you want to, and that’s a really big problem in manufacturers, because if a robot does something bad and breaks the entire line, you can lose millions, and a friend can go there and change a little code and say, okay, that’s not the robot that’s the problem, and we must sue that manufacturer So, blockchain can change that, if we validate a transaction , so, yes, these will be used, and the market will need it, but not right now, in the near future, until the blockchain technology is more stable and there are more standards and interconnectivity among them >> SPEAKER: Thank you. Thank you very much. Okay, so — (Applause ) >> SPEAKER: Hi, everybody. My name is Jason Tran. I’m from the University of Southern California in LA I’m part of the autonomous networks research group, and this is a joint product with Boeing Research and Technology I’m here to present swarm DAG, a Before I begin, um, we have two centers, the center for cyber-physical systems and Internet of things and blockchain at USC, so this is the blockchain-related activity at USC. We do have a hack-a-thon coming in the springtime, February/March, nothing has been planned yet, so I’ll get back to you guys on that Feel free to contact us at [email protected] All right, so, for this work, three things motivated us, cap, base, and EVS. So, originally, the cap theorem specified distributed systems can only achieve no more than two of these three, consistency, availability, and partition tolerance, but then the base philosophy came around by Brewer himself, um, which relaxes consistency to allow for eventual consistency and provides trade-offs, um, for cap. So now, you can trade-off between consistency, availability, and partition tolerance, and lastly, we have extended virtual synchrony, which provides dynamic memberships, which means you can allow partitions to occur and continue to operate as opposed to the typical blockchain tools, which would block if, um, a partition, let’s say a network splits perfectly in half. So, for swarm robotics, the applications need partition tolerance, when partitions occur, each partition should

continue to operate, and they also need a consistent ledger for reconciling conflict So, as swarms, um, partition, decisions may be made where, um, they actually conflict one another, so as long as you have a consistent ledger that is shared, then you can go back in history and resolve those conflicts over time So, swarm DAG is essentially a partition tolerant distribute ledger protocol. We only talk about creating a consistent distributed ledger. As a summary, um, it aims to provide an eventually consistent DAG-based distributed ledger across a swarm, assuming network partitions are not permanent Now, I think this assumption is pretty fair, because swarms will go around and they’ll eventually, um, rejoin, so swarm DAG also explicitly manages the partition memberships through something called the membership management service, and it uses a partition-aware transaction verification policy to control and optimize transactions when swarm members of interest are not around. So, let’s say your current partition is a size five, and you have a task that needs to be done by twenty robots, you can hold off on that, and right now, swarm DAG is under development, so there are a lot of open questions So, a quick way to understand swarm DAG is through an example So, let’s start with a swarm of 20, each partition runs a blockchain with a classical two-thirds BFT consensus protocol, confirmed transactions are transactions appended to the ledger within the partition, but may not be disseminated to the rest of the swarm, and, um, assuming that the swarm that faces a network partition eventually reconnects, finalized transactions are transactions that will eventually be shared with the entire swarm, and the swarm DAG ledger itself is the union of all, only all of the, um, finalized transactions in the swarm at anytime So, you start with a swarm of 20, you start a network, you have a genesis block, the network creates transactions, and then the network splits up in half, and at this point, if you’re using classical two-thirds, it’ll block, so nothing will happen, so what swarm DAG will do is, um, through the membership management service, it’ll detect, um, a partition change and then create two new partitions, orange and blue, and for each partition, a new instance of, um, a blockchain network gets created, so two new genesis blocks They continue to make transactions, and eventually, they move close enough together to be in contact, and they, um, update their parition to include the entire swarm, which is why they’re back to gray, which is the original color, and after that, a merge block occurs, which has two parent blocks, and, um, everything is reconstructed, so now there’s no genesis block, it just gets pointed up towards the original, um, blockchain So, as a full swarm, they create transactions again, but then 15 robots go off to the side and leave 5 behind, so those 15 robots actually can continue to operate under two-thirds BFT, but the 5 robots cannot, so they immediately realize no transactions can be made, so they create a new partition The 15 robots will start to create new transactions, two more, and the reason why they’re dashed here is because they’re not finalized. These are confirmed within the ledger, but these are not finalized transactions So, essentially, those, um, those two transactions will be pruned So, after awhile, the, um, partition of 15 robots realizes that, hey, we lost 5 members, so, um, they update their partition, and now they, um, have 15 in total, and it’s purple, and so these two new partitions move forward, just like before, they create transactions, and eventually, um, the larger partition gets closer to the smaller one, and the same process happens again, a merge block occurs, and then the reattachment up at the two genesis blocks occurs, so up here, and, um, essentially, again, these two blocks are just pruned, so they’re not part of the consensus or the finalized, um, ledger So, the membership management service, um, runs on all nodes to enable and capture these partition changes, just as we showed in the example. It is essentially a modified version of the total membership protocol, which is sort of an older paper So, um, we use that same logic, which uses single ring consensus a little slow, but this is just for memberships. So, each node will broadcast its presence and gossip the current peer list,

and all these gossip messages that are received at each and every node will be tabulated, and by doing so, um, each node will continue to detect if there’s a change in a membership view. If there is, a node will propose, um, that a membership change occurs, and this should be done periodically, not event-based, so as to reduce the network traffic, and to minimize the impact of the membership changes, robots in general should not consider the last tab blocks. If you guys remember the non-finalized transactions, so you shouldn’t take the last blocks to be finalized at all, just wait on those, and so that, um, the policy for that is still under development for swarm DAG, and we’re still looking into it, and earlier, I talked about the partition-aware transaction verification policy, and here’s a block diagram to explain that If you start on the left, a new transaction comes in, you go to the PTV policy, and this is where you decide, hey, is there enough robots in this swarm, or are the robots that I need in this partition If it’s not, then you move over to pending due to transaction Otherwise, if you’re ready to make this transaction, it can be confirmed on the single partitions ledger Now, when, on the right side, a membership update may occur, and at that point, if a merge detection occurs, then you’ll create a merge block, that’s why we go straight to the swarm DAG ledger being finalized with the merge block, and this will trigger, um, each and every node’s pending due to partition queue, to see whether it’s time to try to commit this transaction. Okay, so this, um, picture right here, I’m not going to go into it, but this is in the paper, so go ahead and take a look, and it also describes it in the paper, so I gave you guys a simpler explanation. So, this is not presented in the paper We’ve already started trying to implement, um, swarm DAG, and everything is, um, implemented in containers. We use tinderment for a single partition network instance, we use a patchy tingerrograph to store the finalized transactions This is pretty lightweight, so this might be a good idea for containers We use the gremlin query language for tingerrograph, and we’re exploring lid P2P, which includes gossip messaging for the membership management service, and lastly, we use blockied to emulate partitions and shape the network traffic, and this is a tool from something called the chaos engineering family of tools. Netflix made this really popular. They essentially introduced failures into their network to, um, make sure it’s hardened Um, okay, so, here is a block diagram of the current implementation and test net So, essentially, um, blockade is a command line interface It, itself, will docker run containers, it will, um, use the TC command line tool inside each container, as well as IP tables, to, um, emulate partitions and shape the traffic, and at the beginning, it needs a blockade.yamal, which gives it all the information of the containers it needs to start itself, so you could write a python script to coordinate everything, or you can run a shell script, but at the end of the day, right now, blockade is only a command line interface. We have, of course, a make file to build and run containers, or build and run our entire, um, test net, and, so, um, like I said, we’re using dock containers, which creates a bridge interface, with a and so all the containers are on the same subnet. Inside the container, you can see we have swarm DAG, which is a tinderment app, and it talks to the tinderment core, which is a single binary that runs through the ACI or blockchain interfaceface, and it also runs the membership management surface, which we’ll be using lip P2P to gossip across this interface, and also, um, whenever transactions are finalized, it’ll use the gremlin query language to then interact or store transactions into the tinkerrograph, and, of course, the last line right here is tinderment itself actually runs gossip under the hood, so it’ll be gossipping across the bridge interface as well. So, I think that’s all I have. Yes Um, so, yeah, saved a little time there (Applause ) >> SPEAKER: Questions?

>> SPEAKER: Resilience to members of the swarm being destroyed, like if there’s a partition and those partitioned ones never come back >> SPEAKER: So, the way it works right now, if they don’t come back, then they just never remerge, and as long as, um, let’s say you have a task, as long as there are enough robots for your task, eventually, you should be able to do it. So, if you want the entire swarm to do something, um, that’ll, you might have to create some semantics, some rules. Say I’m going to try to get the whole swarm, but at least two-thirds of the swarm needs to be present for us to act on something (Off mic.) >> SPEAKER: We’re trying to de-couple that. We don’t want to put, because, yeah, swarms can be lost or just go down, so we don’t want to account for each and every swarm member member >> SPEAKER: Can you tell us a little bit more about lip P2P? Did you try to load it for the high loads, and what capacity of the messaging inside this protocol? >> SPEAKER: We’re still trying to figure out how to use the library. It is in its early stages. I believe even the, um, ethereium folks are considering it for its next generation of gossip messaging, so I can’t comment much on that It’s definitely in the alpha phase, but it’s called gossip sub under lip P2P >> SPEAKER: Um, what are your ideas for future work in terms of, like, experiments? So, do you want to test how your algorithm behaves, for example, in long-term, like, experiments, for example, hours or even simulated days, or you’re thinking about, like, overhead, how to minimize the overhead in the messages that you’re producing? What are your ideas? >> SPEAKER: So, in terms of minimizing overhead, it’s a little tough, because we’re an academic environment, we don’t have a huge team that can create really optimized code, so we’re kind of staying away from that for now, but we want to use blockade to really emulate a Barack of partitioning to see if our algorithms are, in fact, consistent over time. So, I believe we should be able to at least emulate about 50 containers on our server If we need more, we’ll just contact AWS >> SPEAKER: Okay, thank you very much (Applause ) >> SPEAKER: Good afternoon My name is Vitaly, and I’m going to continue speaking about robonomics now, and I’m going to tell you why and how you can use robonomics in smart city applications. First of all, did you know that more than 80 percent of people in developed countries already live in cities? And that’s fresh on logistics networks, transportation systems, things like pack package delivery Robots are actually the most flexible and fastest way to deliver, and people found out about that, so there are already 23 bill billion connected devices, and this number is expected to grow to 75 billion, and a bunch of cities already announced their smart city progress. Those are mobile robots, like drones and autonomous cars, sensor networks that deliver data so that cities can make actionable decisions, and industrial robots that responds to human needs Altogether, they’re called cyber-physical systems. Um, now, people, whenever people need to manage, um, a network of cyber-physical systems, they use centralized solution, so there is a central data processing system that collects the data and makes the decisions, but oftentimes, as the number of transactions grows, it results in inefficiencies and actually becomes expensive, and besides, when we talk about smart cities, there are risks with privacy and security of such a centralized solution. So, I don’t know about you, but I don’t want to imagine a situation where NIA goes crazy and gets control over a central center that controls all the robots So, that is why robonomics proposes a decentralized management system for robotic systems It includes privacy as the data is stored and also improves

security of the system overall. One of the key things about distributed ledger is smart countries, and for the first time, they actually allowed us to combine technical and economic details of the transaction into a single instrument. So now, whenever you create a contract with a robot, you can actually, um, be sure that the robot delivers the mission accordingly. So now, I’m going to show you a few use cases we have built. So, the first one was ordering an autonomous taxi service, where a client, um, we showed it on a demonstration, and the client sent a demand for an autonomous text to service, it automatically matched the supply and demand and created a smart contract so that the autonomous car can go on and deliver the mission and move over that small smart city. The second use case we built was environmental inspection with drones. So, we worked with a company that does, um, that works on natural capital markets and does, um, makes it more transparent, so they needed a way to collect data in a more flexible way, so we built an autonomous fleet of drones with the sensors onboard that can actually take off automatically based on the request from the autoimator and then go on and do an environmental inspection The drone actually signs the data, publishes it so that it is, um, nobody can actually interfere with the data, and the operator can trust it The last use case we did was wildfire So, we worked with emergency services to use drones to identify wildfires It is actually possible to use a swarm of drone to cover a large area of forestry to identify fires and, um, reduce their response time for the emergency services services So, thank you very much for your attention (Applause ) >> SPEAKER: Any questions from the audience? >> SPEAKER: The other speakers can chime in as well, because you mentioned the idea of a smart contract relaying to blockchain, but you can very easily have smart contracts outside of blockchain If you think about the fact that you can just, or that smart contracts that are defined by this self-enforcement, any kind of system that is going to allow this self-enforcement can lead to the same results, so what exactly is necessary for you from a smart contract perspective, like a system perspective, from blockchain that you could not create otherwise, like the same question that I asked earlier? >> SPEAKER: Thank you for your question. Those were two separate things. So, blockchain allows us to talk about security and privacy, so as, um, people, like, services in a smart city would not store data in a central cloud, which is a huge privacy risk, and then they would store it on their local, um, devices, but then smart contracts is kind of a separate idea from that, where it allows to actually create a contract with a robot directly. So, unlike the situation where we trade a contract with a company, they, for example, and then the code was not run, something didn’t happen, it’s disconnected, a smart contract actually allows to combine those two, so that whenever you create a contract with a robot directly, you know that it goes and delivers a mission. So, blockchain and smart contracts, yeah >> SPEAKER: But the fact that you create a contract with a company and that you were mentioning something, that the code doesn’t get executed or something like that, that’s because of the fact that the company is not a technology, or it’s not a physical technology, but if you have, um, I don’t know, you can create smart contracts, um, with bank accounts that don’t use crypto currencies, and you have exactly the same amount as you might have with a piece of code that says if these conditions are met, then these results need to be, um, invisaged. You see what I mean? >> SPEAKER: For sure, and that’s exactly the benefits that smart contracts bring That’s not related to blockchain in particular So, that was two separate slides Thank you. Are there any other questions? >> SPEAKER: Anymore questions from the audience? I have one question. So, are you planning, like, to create,

like, a token, like for drone employee, are you thinking, like, about this? No? What are your thoughts? >> SPEAKER: So, yeah, we’re now using a robonomics network, which has a token which we can use for transactions by autonomous robots, and that’s what we plan to use, and a drone employee is a set of solutions, specific solutions, like environmental inspections, wild wildfire monitoring, which we work with clients directly and then use the network to actually bring those >> SPEAKER: So, are you thinking about, like, prioritizing the service, like, for example, to governments, to city halls, or you are thinking more about, like, the final user, like the person that is at his home and wants to get services, like from a drone, for example? >> SPEAKER: For now, we’re mostly focused on, um, enterprise and government implications, so that’s where we see the most of demand for that, and that’s the most capabilities for us to execute by projects and real use cases So, for now, it’s commercial >> SPEAKER: Okay. Thank you very much >> SPEAKER: Thank you (Applause.) >> SPEAKER: So, I’m going to present robotchain So, the current situation in factory environments is we are offering robots and various types of robots For example, the first one is an AVG that transports materials on the factory, and the second type is a rout arobottic arm, and these types of robots produce various types of information, and they came from multiple manufacturers and may have different interfaces and ways to communicate. So, what is robotchain? It’s going to be a private blockchain that is industry-oriented, meaning oriented to be used in a factory environment, that is capable for multiple types of robot data, so any types of robots may be integrated on to the blockchain, and it’s also capable of running smart contracts to interface with the robots operating on the assembly line or placed in the factory We have this model concept. We have the robots, the robots are going to be connected to a compute module, some form of device, and that device connects to the robotchain itself, or is part of it itself We want to separate compute devices so we do not, um, the blockchain does not run and may attempt to remove performance from the robotic line itself and to prevent any other type of situation, and also to serve as a uniform way to connect the robots to the area network. We also want to develop query notes, where the management from the factories can read information from the blockchain itself in order to change robots, to maintenance, and whatever they really need to do They are meant to interface with many robots, that’s since they come from many manufacturers, they need to interface with anything that the robot provides, and with the possibility of smart contracts, that interface must be bi-directional at least in order to effect any change on the robot itself So, about query notes, this is an example from the blockchain explorer

that explores, um, the blocks generated on the blockchain Our attempt is to provide something like this, but in relation to the data we are storing and creating and generating on the factory floor So, we are using this technology as a platform for smart contracts, decentralize the application, and we choose this for two reasons, the formal verification, where we can ensure that the code is, um, going to work, formally using mathematical equations and stuff like that, and it’s self-demanding, meaning that I can do changes on the platform itself, on the blockchain network itself without either losing the previous blockchain information and allowing it to expand further We did the initial validation with blockchain machines where we used the number of nodes and the blocks per cycle This network works in cycles cycles At the end of each cycle, the currency rewards are given, and the proof of stake is reset. So, we did five trials, and our results, um, were this. In five nodes, we got an average of the reaction time from the moment I sent the transaction to the setting where the blockchain says, okay, we are taking care of your transaction, it’s about 1 2nd average time per transaction, and with an increasing number of nodes, we see that, um, the number of seconds are going to 4, which is a lot to what we are trying. We are trying to improve the transaction speed by ten-fold at least. In a factory, robots are always moving, and we need to get everything on to a log We already have a live network running on the laboratory with multitudes of computers entwined, and we are attempting to to connect the laboratory to the network itself We want to create an interface with the blockchain, and that’s it at the moment These are the references for the information used on the presentation, and thank you. If you have any questions — (Applause ) >> SPEAKER: Just not a question, but, um, a comment We already realize the communication for the blockchain and the robot robotic system is open source, you can actually use it, but, um, did you try, um, another chain, maybe something like hyperledger? >> SPEAKER: We simply don’t, I started to do the work in September, I didn’t have much time to investigate every possibility, but we are still open to that, even though this is a foundation project, a funded project, so it’s not entirely advisable to test another application >> SPEAKER: So, okay, I completely understand the motivation of your work, so what are you trying, like, to, so, how do you envision, like, this technology being applied? What is the perfect scenario in which your technology makes the most sense? >> SPEAKER: I had this conversation with a person that works in an automotive factory, and it is a simple conversation, so I don’t have any basis for this, but the thing is, they had ATV that was self-constructed by the factory, that another ATV that was from a manufacturer, they had two robotic cars from another manufacturer, and the conversation was something along the line of we are getting too many persons, and those robots were managed, no, repaired by the tool

manufacturers that they provided, and it’s getting too much robots in the same factory, and we need to have accountability for this If something, if two robots are going to interact from different manufacturers, um, something may happen , and we need to know who was at fault, if it was robot A or robot B The thing with using blockchain technology is the idea of I can say it’s this person, in this case, this robot that was at fault without any person changing the logs, for example I could go to the robot and change the database or file or whatever, and, okay, it’s not my robot, it’s some other robot’s fault So, that’s what we are trying to attempt with the blockchain So, I think it’s a very interesting opportunity for industry, not tomorrow, but in a few years, I hope >> SPEAKER: So, for example, together with these, do you think that we need a new set of tools, for example, in order to search those logs, in order to get knowledge from those logs? So, do you think that besides, like, your infrastructure, you will need a Google of, like, these blocks in order to get information out? >> SPEAKER: I assume so For example, AI technology for searching, instead of having a human searching for thousands and thousands of blocks and entries on the blockchain, using AI to search the blocks and see, for example, potential points where this robot is not functioning like it was working two months ago, for example, so that robot may need something new, maintenance, or it’s simply getting old, something like that The idea was to present to higher-ups on the factory or enterprise system to see the information on a blockchain in an easy way, not the complicated way that we programmers are used to seeing >> SPEAKER: I completely agree with you. One of the things that people don’t realize is not only about this transaction theme, but also the meta information, right? So, what can you infer, like, through processing this, like, meta information? As you said, for example, you could say, well, because of this behavior, So, thank you very much (Applause ) >> SPEAKER: Okay, so, with this, we finished the second session of the day, so all the papers are done by now, so we have a little bit, like, time for some coffee. 3:30, we’re going to have a demo, like for the vending machine of the current initiative, so, libe like, if you want to check it out and touch it, this is the, it will be, like, a nice moment, and yeah, so, see you at 3:30 See you (Applause ) (Break Taken ) >> SPEAKER: Hello, everyone Nice to meet you. I represent

the digital currency initiative, and I want to talk about a

couple of things things, about what we’ve done with the

vending machine and a couple of things about the agency

state of the blockchain ecosystem, and a short

introduction to the lightening network. So, we got a little

bit of a good amount of ground to cover. Let’s start with the first one, um, which would be our mission, right? So, what does the DCI do? How many of you, have you heard of us? Okay, decent. So, we’re a small group that focuses on the, um, research of crypto currencies. Our mission is to empower individuals to make it as fast and easy to move value across the world as it is to move information. So, basically, empower individuals A couple of things about the team, Neha is the director, and you’ll see her right after I present here. There are a bunch of other folks too I’ll point out — because he co-invented the lightening network I’ll also point out — because he helped us develop a lot of the solutions here. That’s it With regards to our advisors, you know, these are the folks that help us the most. They’re very involved. I would also encourage you to read Michael Casey’s column today, the one from today is actually pretty good, and I think it’s very timely. It covers the, um, the crisis that we’re kind of seeing right now with crypto currencies. So, these are the folks that helped us build all these machines The one person that I wish I had here is Jonathan Harvey, who actually did the original vending machine in 2013. By that, I mean he actually made it work with bit coin He’s the one that got it to a stage where we could use it

However, if you’ve transacted with bit coin, you realize it doesn’t function very well for face-to-face commerce, and that’s where I think we tried to address that particular with the, um, with our solution, if you will So, these are projects that the DCI works on today A lot of our challenges, basically, is move information and research that happens within our, um, within our team to other teams that are outside the media lab most of the time, right, and we do that with this concept of working groups These are all working groups, you know, basically, we take a problem, we apply a cross-functional team to it, and then we see how exactly would we solve that problem together with our members, and this particular case, one of the solutions that we work on a related to technology, because it also applies with, it aligns very well with our mission. Um, this is, these are some of the things that were done in the past, and, um, I would say it highlights some of the research we’re doing We teach classes, we, um, we make them all available online, they’re all on our GitHub, we organize events The one that we really liked recently was the L2 Summit, which actually happened here at the media lab We developed bit coin core, so we hired two core developers, and together, they, you know, they developed four teams in the last, no, the bit coin core in the last three years. You know, we do a bunch of other things as well We’re recently starting looking atvulnerrability, so we found two so far, and those are kind of the bigger ones, but we don’t want to do that full-time. It is something that we think is very important. Okay, so, um, this is kind of the agenda Let me start with the blockchain ecosystem, right? So, something that I think I should have said is that we apply a lot of research and thought to public blockchains We do have some, um, we do have some projects that cover permission blockchains, but we, let’s say we prioritize public blockchains, and it’s mostly becausets because that’s kind of an area that has harder challenges, and it’s not very well-established either. So, this is a very interesting, um, I think, um, example that comes from a recent blog post. It’s pretty cool, and it addresses how exactly did the Internet develop, right? One of the interesting debates that I think we’ve heard many, many times over is what comes first, and here, the folks push back and say, look, here’s the thing, it seems that it’s that clean, but it’s actually not that clean Basically, you have this interweaving of, you know, you first think about an application, and then you develop the application to support it, and then, you know, you find another application, then you build another infrastructure for it, right? So, again, not ours, but we think this is a good framework to think about. So, if you take and apply this to crypto currencies, which other folks have done, so we’re just surfacing this here, basically, you see the same thing. You have the bit coin infrastructure in 2000, and from there on, you start looking, we got coin base, we got a theory, the ERC20 token, all the stuff that comes with it, then the application from that, so anyway, you got that balance You build applications that folks want, and you realize, okay, I’ve had problems with this, right? So, this is an interesting way to look at things, right? The reason why I like this particular slide, and again, this is stuff that you can find online, we just put them all together because I think it’s interesting, right? These are all solutions for crypto currency custody, right? Today, you’ve got the exchange custody, in which, you know, you go to coin base, you put your money, coin base holds your money, right? That’s kind of the way it works, and coin base has the custody of your funds. You basically rely on them to hold your private keys, and it’s not just them, all these other exchanges, right? Another model would be the third-party, which is Zappo or others that are doing the same thing, and in this particular case, you basically say, look, here’s a third-party that doesn’t do exchanges, but they can hold my stuff, which is great, right? There’s another third model here which is self-custody, which is the one I personally think is very interesting, because this is something that’s fairly new, and it says, you know what, I’m just going to hold my money myself I’m going to take my private keys, store them however I want to store them, however I choose, but basically, I manage my own money, right? So, technically, I’m my own bank, right? The reason I point this out is in the past three or four decades, we’ve had the, you know, the banking model has existed for a long time, and they hold custody or funds, these, the top two categories basically map very well with the pack, right?

For self-custody, you truly didn’t have a solution Basically, you’d have a wallet that managed your cash. That’s about it. You couldn’t really do digital payments without a bank until bit coin came along The challenge though is if you look at all these solutions, they require, you know, a significant amount of engagement from a user base, and that is not true here, right? So, a lot of the work that we’re going to be doing at the DCI is making sure that we encourage these individuals to get engaged, right? We want to educate them that, hey, this is relevant, this is why you should care, and also build another system, right? Because ultimately, you know, you could have a, you choose any one of these solutions with self-custody,but it’s a solution that’s developed for a hundred people, and in order for to us us to truly achieve our vision and for the currency to really mean something, you need hundreds of millions of users, so I think that level of magnitude needs to be adjusted, right? Um, again, the similar model for the ethereium world, and you see all these different solutions that develop, right, and again, they’re trying to map to the world as we know it, but building on a platform that, let’s just say that is being developed right now, so they’re doing a lot of applications on a platform that’s being developed right now, right? So, the interesting part of the ecosystem in which we we’re working right now today is there are not that many use cases that you can say, hey, you know what, this is a real problem that occurs today, so, yeah, when the sentiment turns sour and things are, you know, people are like, okay, we don’t know what’s going on, um, you start having problems, right? Because there’s not a real use case that you can say, okay, the demand from our real use case is going to act as the threshold So, I think what you see with this particular chart is there are many, many solutions, and there’s a very clear path in here, which is folks are trying to develop a system that exist, are trying to mimic a system that exists today, but with a different, um, architecture. So, all of this is built on ethereium. Bit coin has its own chart, right, same similar solution, so we’ve got folks that focus on security, focus on scaling, and we think this is a very big thing. A big one that I think is kind of making noise right now is backed, basically because it allows for the trading of bit coin and for a few other merchants that have manifested, and they have kind of the resources to bring a lot of adoption to the crypto currency, right? Obviously, payment is a big deal. You see all these solutions for privacy for bit coin, and yeah, this is something that happens today, right? But I think as you see, as you kind of really get engaged with all of this is, um, scalability is a big problem, right? Because you need all of these these folks to actually have a user platform and, A, not pay fees that are super high, and B, you need to have, you know, you need to have prices, right? So, the best way to think about blockchain scalability is, um, right now, every single, um, every single transaction that happens in the system needs to be processed by every single node in the network, right? Which is very, very limiting, right? Probably the best analogy that, well, one of the best analogies that I’ve heard is, think about it, everyone in the world would access the Internet through one single access point, it’s not a great great. It would be very, very cumbersome. Another way to think about it is global consensus, which is what happened right now with bit coin and ethereium and all the other ones, it’s very, very useful, but it’s also limiting, to some extent, right? For example, the bit coin limit is 2 megabytes for every 10 minutes. You could increase that, and folks have tried, but you’re going to include, you know, you’re going to increase a little bit, not a lot, and in order to get 2 billion users, you need to increase a lot. So, this is where layer two, scalability, comes into play By show of hands, how many folks are familiar with the term layer two? So, this is one of the proposals for scaling. Basically, it says I’m going to batch transactions, I’m going to open a channel with a counterparty, I’m going to put a bunch of money in that channel, and then we can exchange however many times we want, and basically, you have an open channel and a closed channel. That is the layer two. It’s not the only solution for scalability, it’s the one that we work on because we think it has the most promise. So, the best way, the way to think about this is kind of cementing the network and maintaining the same level of, um, of security for that network, right? So, the reason why we take this approach is, a researcher co-invented

this concept, so he came up with implementation, it was a lightening network, the paper they came up with, there are a bunch of implementations out there, I think there’s roughly, um, you know, less than a hundred developers in the world that work on it, it’s a very interesting solution, it doesn’t receive a lot of attention, and the DCI, we have our own implementation as well, and this implementation was one that was used in the vending machine too. So, let’s look at, um, so, what are the deployments, right? There are three well well-known, um, for-profit implementations There’s us, which we’re a not-for-profit implementation, but we don’t have nearly as many resources, and that’s it Basically, those are the options, right? The group itself is very well-engaged, there are individuals from all over the world who work on it, and for those of you who have followed the progress, it’s actually fairly lively, right? We have tried to implement it with, um, a number of use cases The vending machine was one of them, it was not our primary use case, and then we said, okay, let’s give this a try This is what brings us to the vending machine The first thing is why, what’s the point, what are we trying to prove here, and the reality is we wanted to show people, the media lab member companies that show up, okay, this is how it would be to transact with crypto currencies, right? For those who have tried before, you know that, okay, it works fairly well, if you transact offline If you’re online, you know, waiting an hour or however much you need to wait in order for a bit coun transaction to become, um, usable, um, it’s not a big problem, right? However, when you’re in-person and you say, okay, I’m going to want to use, I’m going to want to buy a cup of coffee from you, I cannot wait an hour for that transaction to be confirmed, so that’s a problem. Um, that’s the biggest one that I think lightening network scales for bit coin, for transacting, is that immediacy of the transaction There’s another one, that you can make it cheaper, because you’re , right? I would say the big thing that we did is we’re lucky that Jonathan has already done a bunch of work on this vending machine, we just kind of kind of repurposed it We settled our code into this raspberry pie, we edited the connection a little bit and then realized we needed to wait for folks to interact with their nodes, so we actually had to create a mobile app for all this. We also wanted to show one of the things we’ve developed, which is this cross-jentomic swap, so that you could have bit coin and live transaction realtime, and we had that, but we kind of had to do a little bit of work on routing and other stuff, so we’ve done exactly that Um, so, the payment routing, I would say, is probably the, um, the hardest one. So, the payment routing, the way it works is, actually, I have a little bit of slides here If I want to transact with Eduardo here, let’s say, you know, it’s one thing if you and I open up a channel and I say, hey, I know my buddy, we’ll open up a channel here, and I’m going to pay for something I know, right? However, if I’m going to transact with somebody I don’t know, maybe I don’t want to open up a channel with them, and it took me time to understand that. Let’s say I go to the vendor down the street and they say, hey, here’s your coffee, you’ve got to pay me something, in this case, it’s two bit coins, a very expensive coffee (Laughing.) >> SPEAKER: So, I go to the vendor and say, look, I don’t have to pay, that’s kind of a problem, but one of the approaches that we thought of, like, okay, look, I don’t have aen a channel with this vendor, but I have a channel with my buddies, and as long as one of those buddies has a connection with that vendor before, I could pass on a payment that they could forward for me, right? In this case, we’d say, hey, this particular entity, Charlie here, says, yeah, I have a channel with this entity, and they will route that transaction for a small fee to the vendor, and all this happens realtime. So, this was significant, this is something that I think, um, other implementations have done as well, which is, you know, you use, um, one asset, in this case, bit coin, and you may should have, you route a payment through however many channels you have, right? Typically, I think, you know, it almost never goes past three or

four hubs, most of the time, it kind of, that’s kind of the big, um, the big number is around three or four hubs, but this, in fact, lightening network is actually live today on bit coin and has a bunch of capac capacity out there, so folks are using it for one asset, right? I think what’s a little bit different for what we’ve implemented is we actually added multiple assets In this case, it was bit coin-related, or bit coin clones , and the way it worked is you basically route through multiple assets, right? The example here would be same thing, you know, I go to the vendor, I give my money, and in this case, I say, hey, I don’t have your asset, I have my own money in a different asset, so I apply the same problem, I say, hey, go to my buddies and say who can actually, you know, I’ll give you, in this case, um, bit coin, and if you have a channel with this vendor and you have a bit coin node and another node, you can actually pay this entity U.S dollars directly on my behalf, right , and that’s kind of the way it works It’s something that I think is, um, is very useful. It will be, it will continue to be more useful in the future as it develops further. I think the main thing that, um, that this brings up is the fact that you have multi-currencies, right? One of the reasons why multi-currencies matter is if you think of the adoption of crypto currencies, it’s actually a fairly tribal system, right? Folks who like bit coin like bit coin, folks who like ethereium like ether, and if you want to move from one asset to another, typically, it goes back to that custody solution, right? You have to rely on somebody to say, okay, here’s my bit coin and give me ether in return, but you’re going to have to hold something for me. There’s another way to do this in which I don’t need to relinquish custody of my funds, so that’s kind of the interest ing part about this You ask, okay, why would I route these transactions? It’s because you get paid So, there’s a fee, a small fee in this case that, um, anyone who routes payments gets to hold for themselves, and the way these fees are, um, are determined is, you know, if you say I’m going to choose a random fee,ultimately, it’s a market base, so every single entity can route fees, and they’re going to advertise the rate under which they will be routing, and the optimization will be the cheapest route. Okay, so the other thing that I think is interesting is, um, we need the mobile app, we needed a wait for way for folks to interact We’ve developed one We actually had two, one for iOS and one for Android We ended up using mostly iOS, so I think it tells you something, that the folks who visit the media lab are partial to iOS, but we did have one for Android too. The interesting part about this here is how exactly do you decide to implement this particular solution, right? There’s, um, one of the things that I think is interesting about lightening network is that in order for it to work properly, you have to always be online, right? And the way you want to run that node, you could technically run it on a phone, but it requires a very stable network connection, and mobile phones don’t really have a very stable network connection, and they use, um, they use data, which is fairly expensive as well. So, our approach, you run the node on your computer at home, and then you connect, um, you pair through it with a mobile app, and then the mobile app basically just instructs the desktop to make transactions on its behalf Um, okay, so what goes into that, right? So, basically, this is a raspberry pie, it wasn’t a new one, we just adjusted it to make sure it works and did all the nice stitching here This holds our software, which we developed, and from there on, um, we added, in this case, it tells the vending machine that, hey, we’ve got value, and it’s instructed to drop coins. So, it says, hey, you’ve got a bit coin worth four U.S dollars, so drop a coin. The environment was this, right? So, we had three different ones We actually have the big router note out there that would receive payments, and everybody

would connect too. We could have kind of done it, um, in a different way in which we wouldn’t have a router note, and everybody would connect kind of to the vending machine or to each other, but this would remove a little bit of, or remove a little bit of the risk, so we kind of do it the simple way&then the demo notes, they would be stored on a virtual machine, you know, person walks up to us and we say, okay, here’s the node you should use, they pair with it through the mobile app, and thenthey make a payment, and the payment gets through the router note to the vending machine and the goods come out on the other end. I’m not going to go into the challenges, because I want to leave a little bit of time for questions. Some of the steps that I think are very interesting on this one is that, um, obviously, this is one use case for lightening network, for our implementation as well, right? It’s one that I think folks can relate to fairly easily, but there are others. They’ve done an amazing amount of work on introducing smart contracts for bit coin, so we have that one. There are a bunch of other ways in which we can develop this, so I would encourage everyone to check out our GitHub. All of this software is available online. Go check it out, see how it is, and if you want to talk to us, just let us know. We’re fairly friendly So, let me pause right there Any questions that you’d like for me to address? (Applause.) >> SPEAKER: Any questions? Don’t be shy. So, maybe I can start by saying, yeah, over there? >> SPEAKER: Thank you for a nice presentation, and the question is, um, tell us a little bit more, how do you provide the security on the raspberry pie? Because there is a lot of things to do >> SPEAKER: I got you. So, the main thing that I think would have to happen is for a hacker to physically breakdown the machine, right? So, they first have to breakdown the machine, from there on, they would have to enter into our codes, which is not trivial either. They would have to be fairly savvy with our code, right? Even if they, I think the most they can do is basically break the machine and kind of steal what’s in it The code itself, I would say, um, there’s not much to steal either, right? So, like, you know, it’s, um, you’d have to break the encryption that would secure the channel, if you will There’s not much to steal from there. I would say the level of security of the money in the, you know, in the vending machine, of the crypto currency in the vending machine is the same transaction that the bit coin would have as well So, if you were able to extract the key somehow, there’s not much you can do. Probably the biggest problem is breakdown the machine and steal what’s in it >> SPEAKER: Do you think the private keys, um, is okay? >> SPEAKER: Yeah, I mean, the way we’re storing it, they’re never stored on the, um, on the actual mobile phone, they’re only stored in the ras pberry pie, and yeah, we think they’re safe in there >> SPEAKER: Thank you >> SPEAKER: Thanks for the presentation A question on delivering value and also the link to swarm robotics. There is value in labor as well, so that was one link I saw, but I’m curious, do you also interact with, um, the group here, and what is the link you see? >> SPEAKER: That’s a good question, right? Which is, okay, we develop all these solutions, and that’s a good way to create value, but how do you deliver the value? >> SPEAKER: The broad question is how do you see a link? And the link I see is also that robots can be provided to the swarm to deliver a service, and I can own the robot and thereby transfer value >> SPEAKER: Yeah y, I mean, look, the main thing I would say that this solution introduces, um, is we were trying to show the fact that every machine, no matter how old, could actually be retrofitted to accept these payments, you know, realtime, right? One of the big things that I think we know is a problem is if you want to send micro payments, it’s actually very expensive, and, you know, it’s very hard, right? Not impossible, but it’s very hard, and I think one of the

proposed solutions for lightening network is actually fairly useful for that. It allows for micro payments to be much more realizable 6789. I wouldn’t say we’re right there yet, because it’s still kind of early technology, but this is the path forward that I can see, right, which is, um, the only way you can have, um, micro payments, that I can think of, is through something like this, right? Because you still have the underlying security around, okay, how do I make sure that somebody’s not going to steal my money, right? And I think the solution that, in this case, bit coin provides, bit coin, as originally intended, wouldn’t have worked, because it’s expensive and not very easy to use, however, a solution that would allow for small payments, kind of function between robots, it actually would probably be a good way to do it >> SPEAKER: Maybe a follow-up question, if I may. Do you also see the economics of those systems then within your scope? Like, for example, the robots get an incentive, if they perform an action, and, so, you have sort of an evolution in robots, that a few robots will do well or perform actions, they get more incentfs, more tokens — >> SPEAKER: Yeah, that’s an interesting question, right? I think one of the things we learn from doing smart contracts or from looking at them is that they’re very hard to write It takes a lot of, um, it takes a lot of skill and really useful tools to design that ecosystem, right? So, I would think you’d have to have a fairly simple, um, mechanism implemented, and then from there on, you basically have, um, kind of free competition, and you allow for, you know, a different set of robots or different swarms, I would say, to, um, to contribute to that market, basically emulate a marketplace, right? So, it tries to say, okay, kind of bring in mechanism designs from economics and let those economics kind of figure out the way these robots are going to work together, right? I don’t think you can design a system, a perfect system, kind of thinking, you know, like a self, um, a self-reliant system, I think the better way to do it will be to explore what the folks from economics have developed, but that’s the way I would think about it, but I actually don’t know It’s not my area of expertise, but yeah, please go ahead. It seems like it’s a really good, useful area >> SPEAKER: Now I’m thinking as well that the same holds, that if we are able to transfer micro payments across the world, what will that change in economics? >> SPEAKER: Look, it’s one of the things that I think we in the DCI think about a lot as well, which is like o cay, you bring this, kind of bring this tool, right, all of these micro payments are tools, you bring this tool and then say, okay, what do you do with with this tool? By itself, it doesn’t mean anything, it’s what you’re going to do with it, and what’s the reality that you’re encouraging? Is it that, you know, am I going to pay for every single vowel that I’m going to read in an article? Is it possible? Yes, folks have suggested it, it is possible Do I really want to read like that? Maybe not. We kind of have to figure that out, but I think, you know, the tool is going to get there, right? This tool will become better and better, and I think they will be able to support all of that Now, what will folks, you know, do with those tools? I think that’s up for debate, and, you know, how will we decide to implement them, right? >> SPEAKER: O >> SPEAKER: Last question, if I may. Do you only look at monetary currency? >> SPEAKER: So, we’d like to look, you know, in other ways, in other types of kind of non-monetary currencies. Right now, we’re leading with monetary currencies, because we kind of understand that a little bit better, but he has many passions, and one of them is non-monetary currencies, and he’s got a really cool article on that It’s, um, it’s a little bit further away, but I don’t think we’re against it, it’s just not a high priority for us right now, but I think we’re going to get there >> SPEAKER: Thank you for asking one of my questions (Laughing.) >> SPEAKER: How much focused are you in currency with blockchain was one of the questions, um, in general? >> SPEAKER: I would say a lot of the interesting problems right now have something to do with currencies, recognizes right? It’s not the only thing we do, we’re very much interested in privacy as well, to make sure you make these movements, but you don’t want everybody in the world to know you’ve made these movements, so privacy is another one Um, we have other focus, I would say We’re not exclusively on currencies, but it’s very hard

to not touch currencies >> SPEAKER: And how open are you to adopt a partner with other kinds of ideas? >> SPEAKER: I would say we’re trying to kind of walk that thin line about, you know, being open to partnering and collaborating with other ideas while still being somewhat focused on what we’re working on. So, I think the challenge for us is, you know, if we always follow the next, new, shiny thing, there’s always going to be something, so we kind of have to follow our vision. That’s why whenever we start presentations, we start with our mission, because it keeps us grounded into what we’re trying to do. That’s the thing we’re trying to solve for, right? But the way we get there, I think we’re fairly open, right? If I go back to the list of projects, we’re goingz going to see that a lot of them are, um, yeah, you know, I think the name itself, the digital currency initiative, may be a little bit misleading. We do more than that, but, you know, in 2015, when we were established, that was a good name >> SPEAKER: Thank you very much >> SPEAKER: Thanks again There’s many different blockchains, and you talked about the landscape and how it can change unpredictably How do you think the solution might change if some chains have full blocks and some chains are empty? Do you have an opinion on how that’ll change over time? And what if there are some successful chains that maintain, um, blocks that are not full? >> SPEAKER: That’s a good question, right? The way I understand the question is are we in a multi-con world forever, or are we going to have one to rule them all? And reality is I think we’ve tried the one to rule them all approach, and really, what the ethereium platform showed is that folks do want to try other coins. One of my colleagues likes to say that money itself is a very social construct as well, it has a very big social component to it, and, you know, to some extent, you know, if I see it replaced by somebody else, it doesn’t necessarily mean I want to join them either. I think the way we’re thinking about it is we’d like to make it easy for these currencies to interact with one another and let each of them and their communities decide whether they’re useful or not So, you know, to some extent, a lot of of folks have kind of had different illustrations about different blockchains and whatnot in terms of, you know, how big their throughput is and how big the demand is as well, right? It varies significantly I think that’s probably one of the problems, when the use case you’re pushing for is speculation, is that the demand and kind of supply of currency is going to vary widely, and that’s one of the problems we’re seeing today. I think the way we think about it is there’s room for more than one coin, for sure Will it be thousands? Probably not Part of the reason we’re biased towards this multi-currency world is we think it becomes unwieldy after a certain time to convince everybody to use one particular coin, because that is the right one >> SPEAKER: Well, let’s thank the speaker once more (Applause ) >> SPEAKER: Now, I would like to invite Neha and Thomas to join me. We are going to have, like, a panel discussion for the last session of this event. A very relaxed panel discussion, I will say So, maybe, so, you know Thomas from before, right, but maybe I should introduce Neha, the director of the digital currency initiative She’s an expert in blockchain, she’s actually the expert in blockchain here in the media lab So, this final panel discussion has basically two objectives The first one is a little bit, like, to be able to have two experts in the matter here, so we can talk to them about what do they think about these, like, new synergies of blockchain with other stuff,

rather than crypto currencies, for example, robots, IOT, etc., etc , and the other is to reflect on what we’ve seen since, like, 9:00 a.m. I’m going to start by throwing some questions back to them, but after that, please, you know, don’t rezitate hesitate to ask whatever you want. This is very rare, to have people that really know about these topics, so I encourage you to ask whatever you think. So, the first thing is, like, I want to ask Neha, so, you’re an expert, like, in blockchain technology, you have been talking about the goods and bads, the potential of this technology and the limitations of this technology, right? So, what do you think about, for example, other uses of blockchain technology besides, like, crypto currency? So, do you think they are viable? Do you think they’re not viable? Do you think there’s some kind of use cases that makes sense? What do you think? >> SPEAKER: Great question, and I’m very happy to be here today with all of you I’m going to push back a little bit on the expert. You said that a few time. You know, this space is so new, I think we’re all figuring it out, and in particular, I don’t want anyone to feel like they’re late to the game, if they don’t particularly know what’s going on. I started in this space about three years ago, so it’s hard to call yourself an expert when you’ve only been doing it for three years. So, it’s so new and changing so fast that I think there’s an opportunity for everyone. Um, regarding, you know, the currency versus non-currency sort of division between things, um, my opinion has actually evolved over time. My background is in databases and distributed systems, I did that here at MIT before I started in this role, and I was initially drawn to the technology, in part because people saw non-currency-related use cases for it as well. I think bit coin, you know, the first crypto currency and the thing that really kick-started all of this had its origins in a certain type of philosophy and a certain type of sort of thinking about, you know, money and what it means in the world, and I still think, you know, there might still be something to that, but, you know, as time has passed, I’ve sort of started to view this just more as a pure technology. However, we should never underestimate the power that money has in our world You know, the more you learn about money, the more complex you find it is, and the more you realize how much of what we do is driven by it. You know, the monetary aspect, I think, makes this technology more powerful, not less. So, that said, I’ve gone, you know, back and forth a million times, um, I think from a pure sort of what is new here sort of question, um, the new part is using incentives in order to reach agreement in an unpermissioned network, in a permissionless network. In order to do that, you require something with value. That, to me, is sort of the new research angle here, was being able to reach agreement in a permissionless network Um, that said, that doesn’t mean that just because, you know, one might argue blockchains, as we’re talking about them now, are sort of distributed database technology that we knew how to do in the late 80s, that doesn’t mean that, you know, their time hasn’t come, that we aren’t seeing people think about them in new ways and talk about them in new ways, because quite simply, we haven’t seen multi-organizational databases before, and I do think multi-organization databases, or if you want to call them distributed ledgers, it’s a very powerful concept >> SPEAKER: Okay. What about you, Thomas? >> SPEAKER: Um, so, some folks might notice that certain, a large vendor based in New York, um, they have been focusing on supply chains, right? So, blockchain systems, typically, people talk about applications, it’s usually digital currency or supply chain. Now, given the fact that the U.S. regulatory authorities have not really made-up their minds about, you know, whether crypto currencies is an asset or whatever, um, there’s definite uses of blockchain for supply chain I like to say that a lot of the problems that we have today in the world is the supply chain of paperwork, it’s the supply chain of contracts, right, whether it’s buying and selling shares or whether it’s a true supply chain of goods, you know, Boeing getting parts from their suppliers, and a good example is the IBM example Mersic is a big shipping container company, and their work flow of how to do shipping hasn’t really changed since the 60s and 70s, it’s still paperwork, and one of the problems they have is

that when, um, paperwork gets handed over, literally paperwork in these plastic sort of folders get handed along, sometimes, at certain points in the supply chain, the physical supply chain, people make changes, so the truck driver has to deliver this container into a country, he’s at the border, he has a timeline, you know, to clock his ticket, so he will, you know, modify stuff like that So, that’s one aspect, so supply chain provides more accurate, blockchain provides supply chain with more accurate visibility into shared information, number one. Number two, because you have this shared visibility, there’s an opportunity for people to optimize, um, their capital use Instead of having, you know, um, your warehouse be, you know, always full with inventory, if you could see two or three, four steps down the supply chain as to who’s consuming what down there, you can adjust your warehouse levels considerably, and that means that you free up capital, rather than having a warehouse full of inventory and having tens of millions of dollars sitting idle for three days to a week You can either free up capital or even collat collateralize that. Number three, the use of blockchain in supply chain scenarios allows for more efficient, um, what is it, of a payment, and I have to put payment in quotes, meaning that settlements of buying and selling on the block chain can be done quickly with an in-chain tokennized payment system, and how that gets converted is at the edges, either at the beginning or the end, but if you are a supplier of components for a big company like Boeing, you need to get paid quickly, right? Assuming you send it and there’s evidence this thing is on the way to Boeing headquarters, you know, you want to be paid, because you don’t want to sit around for a week, and what if you could have an in-chain token, and then you can pay your other suppliers upstream for the components that make-up your parts? So, I think that’s the two things So, definitely, um, supply chain currency, again, it’s, I think, a more difficult question, because it deals directly with, um, our economy Supply chain is definitely a great match and something people could do today, people are doing stuff today with supply chains >> SPEAKER: Great. I actually have to say two things The first thing is it’s very interesting that you mentioned the word incentive, because this is what I’m trying to work these days I’m trying to make this gain theoretical approach to these swarms of robots in which, of course, there’s going to be robots that are, like, malicious or have, like, problems, and at some point in time, they might, like, provide bad information, like, to the group, but what we’re trying to model is the fact that, actually, if you lie to a group, this is going to cost you money, right? The fact that, for example, um, if you provide some information to the group and then you put, like, you use ether or, like, bit coin and you put some escrow in that information that has to be checked, like, at three blocks afterwards or something, at some point in time, if you lie, that money will never go back to you, so at the end, you will not be able to speak, right? So, this is a very interesting way in order to, like, model these incentives, like not only in groups of people, but groups that need to run autonomously for a really long time, right? One of the things that impressed me, like, during this event is the fact that many people are talking about permission blockchains, and also how to log things, not how to search for logs, no? So, we have, like, a factory in which, like, robots are from different vendors or different devices are storing information, and at some point in time, something happens, and you need to open that, like, black book of, like, automation, right, and you need to decipher who did what. So, it’s very interesting that, like, not one paper, but two, three papers actually mentioned this thing. So So, do you see the future of these, like, systems more into the permission, like, blockchain world? Do you think that’s the entry point, like, to this, or do you think that, for example, public, like, blockchains have higher impact, more towards, like, a mainstream option? >> SPEAKER: I see a role for both. At the moment, I think public blockchains are sort of the thing that we find most interesting and where we’re seeing the most innovation in terms of applying new cryptographic primitives, applying new economic models, sort of what you just described was staking, right, and it’s this

idea that you lock up some of your funds, and your funds are locked up, and if someone can provide a fraud proof, prove that you lied or that you double-signed or did something, you have to be able to sort of put your problem inside of this box, construct it in such a way, then by presenting that fraud proof, you know, you lose your staked funds, and this is a very powerful concept, right? And the place where we’re seeing it deployed right now is when thinking about consensus protocols for public blockchains. So, um, I think the public blockchain world right now, there’s a lot of money there, there’s a lot of people there who are excited about the vision, and, so, there’s a lot happening very fast. That said, I still think this technology, um, will come to play a role in enterprise You know, it’s usually, sort of the innovation is happening around the edges, I think mostly, and then kind of comes into enterprise, not necessarily comes out of enterprise, but it can happen in different ways ways >> SPEAKER: What do you think, Thomas? >> SPEAKER: As I mentioned this morning, we have both, and we have some issues in having two blockchains, but in terms of permissioned blockchain, we need a finer definition of what permissioned means. So, at the very least, I could see two things So, do I need authorization to read from transactions that are already in the past, that are already confirmed? So, in other words, to even touch the information on a nodedo I need authorization? That’s one model. Another model would be that you have a blockchain where it’s essentially something like, um, basically, it’s permission, but certain transactions are encrypted in such a way for confidentiality, so, yes, you could see a block of public, public, readable, readable, and then a block that’s encrypted, and in order for me to read that, I need some kind of permission. So, there’s different granularities About two, three years ago, we wrote about the possibility of mining nodes being selected, could a mining node look at a whole bunch of end process transactions, say, you know, I want to process this, I want to process that one, I’m going to drop the other one So, in essence, there’s some degree of filtering that a mining node could do based on some authorization sort of model. This is why the word permission blockchain is just too granular I think it’s kind of meaningless right now >> SPEAKER: So, now that we were talking about, like, enterprises, companies, Thomas, can you explain us a little bit more, like, about this concept of, like, bringing the algorithm? >> SPEAKER: Sure So, the OPA work is related to the GDPR, and the folks in the United States spoke up on May 22nd of this year to discover that this four-letter word is coming to the vocabulary of the United States, which is good We’ve been yelling about it for the last two years in the U.S , and most lawyers that we spoke to said GDP watch 2011, the work produced a very important report titled personal data and asset class. It’s worth reading. It’s a list of gripes about personal data, how things are being mishandled and so on and so on It’s a basic flip of the way people do business So, instead of, um, what people do today, so for example, companies like Ecofax in the United States, the reason they’re able to compute our credit scores is because the banks feed them information, so Bank of America will give them my banking transactions, so when you guys open a bank account, you’ll receive in the mail a little booklet with fine print that you can read, and in there, there will be a paragraph that says the bank has the rights to share your information with credit rating agencies So, in OPA, instead of doing that, we send algorithm to the data repository. So, number one, data must never leave the repository, stays where it is Number three, the algorithm must be vetted, vetted to be fair in terms of machine learning, for those who know what that is, because there’s always in-built bias in datasets and also in algorithms, and option number four, you heard Sandy talk this morning about, you know, doing computation on encrypted data, that’s kind of the final stage, but that model fits in nicely, like, with the

supply chain model, because a lot of supply chain companies, they don’t necessarily want to reveal, you know, everything about themselves, they’d like to have some, they would like to have standard APIs, standard transaction format format, and so the OPA model also fits nicely with the supply chain >> SPEAKER: I’d just like to ask you, Neha, what are you working on in this moment? The DCI has extensive work in crypto currencies, in tools for analyzing and building blockchains, but if we could get, like, a picture of the future, what would it be? >> SPEAKER: Sure. Just checking, in a lien stock, did he have a slide up that had our projects on it? Yes? So, that’s a subset of our projects, but I want to talk sort of about the pillars of what we’re working on. So, I think we’re working to address some of the fundamental problems underlying blockchain technology as it exists today. So, one of the major problems is scalability, is how do we actually get this technology to scale to a billion people, because right now, it quite simply can’t, and I think Aline presented some of that our work in that area, figuring out how to use the blockchain as an anchor of trust instead of performing every step of every computation on every node in the network. The second big area we’re working on is actually around privacy, but around zero knowledge proofs in particular So, zero knowledge proofs, I don’t know if you guys have seen them today or not, the idea is that you can present a short string that convinces you that I know something without actually revealing anything about what that is. So, I know the way to factor this very large prime number, or I know the discreet log of this curve point or something like that So, we’re working on zero knowledge proofs from two different angles, so one of them is my colleague who is a co-author of, um, something called ZK snarks, which are used quite extensively in something called Z cash, of which he’s a co-founder, and also, um, in a lot of other crypto currency projects. They’re pretty useful. He’s working on sort of the next generation of those that have slightly better properties. So, that sort of fundamental theoretical work around zero knowledge proofs, but also providing these tools as primitives for projects to use, and then we also had a project called ZK ledger. So, ZK ledger was a project that we came up with, and what it is is it’s a ledger, permission blockchain ledger, um, in which, um, transactions are completely private, you can’t see the amounts, you can’t even see who’s transacting, you can’t see any information about the transaction other than a transaction happened, and, um, the ledger provides provably correct auditing to a third-party. So, let me explain a little bit about what that means. It means you can run analytics on this ledger without needing to reveal the insides of the transactions So, um, we sort of implemented this as a use case for banks, trading assets back and forth, and we had a student apply the ideas of this technology to the asset-backed securities market, so she was kind of looking at, okay, if we had something like ZK ledger, how would this change this market and the way that this technology is done today. So, we presented that work at a conference earlier this year, and we’re in the middle of trying to release a library of the zero knowledge proofs primitives that we used in that project. So, um, look out for that, that’ll be on our GitHub pretty soon. Then in addition to that, we have, um, a few other projects that we’re working on. There’s quite a bit, we’re probably doing too much, I don’t want to list all of it, but I will say, um, one thing that I’m really interested in right now is understanding, um, rational consensus, so things like proof of work, proof of stake, we don’t really understand exactly why these things work or understand under what circumstances they might not work. There’s been some really interesting papers by economists coming out and sort of applying their models to proof of work I think, you know, we think that there’s another angle to sort of take around there, and then, eventually, I hope that we can sort of step into the proof of stake world as well, but we’re not really doing that yet, but we’re probably going to be hiring a researcher in that position, so if you’re interested or know people who are interested, let us know >> SPEAKER: Super interesting This is very related to one of the things we’ve seen in the symposium, that, um, we are moving, like, from, like, um, seeing the blockchain as, um, like, a set of atomic transactions to seeing the meta data, what can we figure out, like, through basically understanding the history of the ledger and see the possible behaviors of, like, the players there. So, this is, like, a line of research that I think is going to be really

expanded in the next year Maybe in the next symposium next year, we’ll have some papers about that With this, I’d like to open up the floor for questions >> SPEAKER: I’ve written it down. It has a short intro So, um, we’re sort of renegotiating social contracts, and technology is a large part in this, and what I like about this is that it’s technology, but it evokes many philosophical questions as well, and it enables us to design systems in an eradically new way So , I think it requires quite a large shift rather than an incremental change to capture its full value. So, my question is do you think this is happening, and do you think that’s true? Are we augmenting a system, or are we changing and redesigning systems? >> SPEAKER: They’re looking at me (Laughing.) >> SPEAKER: I need more beer (Laughing.) >> SPEAKER: I see a lot of problems I encounter that it would work if you changed the full ecosystem, and if you do it incrementally, people are disappointed, because the net effect is negative >> SPEAKER: I think what’s really happening is that society’s beginning to realize that data drives everything, so when we talk about the data-driven society, it’s real, it drives everything It drives the decision for recollect, um, urban planners, and we see this in our projects, we have a number of projects with governments, we have a project with the government of Beijing, and it’s literally a project on designing new suburbs, where you put the train station, where you put, you know, things, and people are interested not just in factual or attribute data, they’re very interested in social interactions data. So, one of the big projects we have, it concerns diversity, so the theory goes like this. I’m not an expert in this at all, but let me put it in lay language In order to get the best GDP output from a square half-mile business district, you need diversity of not just businesses, but people there, and diversity doesn’t mean, of course, racial diversity, but diversity of incomes, education, places where they live, and the more diversity you have, the higher chance that you have better GDP output in the next five years for that area So, that’s a good example of data In a sense, the supply change, the sort of paradigm is also trying to get better visibility, get access to shared data, get access to shared view of state of shared things, and, so, um, you know, histor historically, we never have a clean break, and it’s always evolution, it’s always messy, I will guarantee you, it’s always a mess, and even in this blockchain world, it’s going to be messy, and legacy systems are not going to go away There’s too much legacy, too much investment, it’s not going to go away You might put an API in front of it to make it nice and clean, but in the back, it’s still, like, awful, you know, 30-year-old technology. I don’t know if that helps >> SPEAKER: Yeah, I think it’s interesting, you mention government, I’ve worked with them for a bit, and they have quite adicative envision, so if they want to evoke change, they can, and the supply chain will follow them, and the government can do sort of the same. If they want to do something, they can make that system change happen, so, yeah, that aligns well with my thinking at least, that if you want the full impact of distributed tech, you need quite a massive change with a lot of stakeholders to come with you, and then a government, obviously >> SPEAKER: Part of what, um, the question that, um um, that was asked today about incentives, there is a sense of empowerment that you want to give individuals, and we’ve been saying empowerment really comes from getting people access to their own personal data, and that’s step one. So, for example, I could argue with AT&T phone that I’m co-owner of the data that I generate on my

device, and they will argue back, they will send a letter, their lawyers will say, no, AT&T owns it a hundred percent. So, if we really want to have change, we have to enable individuals, empower individuals to make use of their resources, including their personal data, and once you have something like that, you know, can you layer an incentives model on top of that so that they create different interaction models that bring other people into the network? To me, data is truly the new oil I think that’s where we need to start >> SPEAKER: I think it’s a really interesting question you raised, and it’s something that we sort of struggle with ourselves, is how is change going to happen. Is it going to happen by partnering with large companies and governments to do sort of these big projects? Or is it something that’s, you know, sort of more ground-up? And, um, you know, I think if you look at some previous technologies, it’s really happened in different ways. The one we like to sort of look at and evaluate and understand the most is the Internet, because it was, you know, it was fairly recent, we’ve seen sort of what happened, it was a massive change, and if we sort of look at how that happened, I think it really caught a lot of people by surprise. It wasn’t sort of this top-down, you know, a government has decided to make the, in some places, it was, in some places, it was, in a lot of places, it wasn’t. You know, things really grew around the fringes, around the edges, through start-ups, through innovation, through, um, you know, these cultures, and slowly, these things sort of grew and grew and grew, more people started paying attention, large companies realized they had to pay attention, it kind of caught them a little bit by surprise. Um, whether that’s going to be repeated or not is sort of up in the air. I think the reason a lot of large companies are being so forward-thinking about creating blockchain innovation labs now is because they look at what happened with the Internet or with cloud technology, and they say, okay, I got to get ahead of this thing, but I think I think in a lot of ways, they don’t really know how, because we’re still developing it, we’re still figuring out exactly what it is, and it’s very interesting, I think it took a long time for, probably the best thing the government did in the United States, at least for the Internet, was to sort of give it space. So, instead of regulating it early or sort of, you know, jumping on it early, they sort of just said we’re not going to, for example, um, they didn’t impose taxes on goods sold over the Internet, so for a long time, you know, it gave space for e-commerce to develop, because they didn’t have to pay the same type of tax, whether that’s a good thing or a bad thing, but really, the strategy was just giving it space, and it wasn’t until very recently, I would say, 2008, when we started seeing these governmental open data initiatives, like let’s put stuff on the Internet, let’s actually be more friendly and have APIs and that sort of thing. So, I think it took quite awhile for them to sort of come around to the potential of this technology. In astonia, a whole different situation, so really depends (Off mic.) >> SPEAKER: Isn’t that where the entire centralization creep came about from the private sector? And this is where you have all of these big corporations E-commerce led to Amazon, which is now the biggest, like the mammoth of Internet companies, and, so, isn’t there also the danger of repeating this paradigm, if you say, well, we’re just going to leave, we’re going to keep our hands off the theblock innovation sphere, and then there’s going to be another centralization creep that is going to follow the money. What do you think about that? >> SPEAKER: That is a very good point, and I’m going to push back a little bit, because I do think regulation has a huge role to play in the centralization we see right now, especially in the financial system. If you look at the burdens and overheads to starting a financial institution, it’s massive, and I’m actually quite worried about that happening in the crypto currency space. You know, they’re starting to sort of apply a lot of regulation to crypto exchanges. Overall, I mean, I’m not an anti-regulation person, I believe that we should have market integrity, that we should be regulating our financial exchanges. What I’m worried about is only the top three or four crypto exchanges are big enough and have enough money to comply with regulation, and that this will drive the bottom 90 percent out, sowe’ll end up with centralization at that level that we wouldn’t have necessarily seen if we didn’t have the regulation It’s It’s an interesting question,

whether keeping your hands off entirely forces centralization It’s a different situation with Amazon and the e-commerce stuff, because any company could have done e-commerce, right? It was wasn’t that they created a regulatory regime which made it easier for a large company, I think the large companies and the centralization came about for other reasons, a lot having to do with collecting massive amounts of data >> SPEAKER: Okay. Any other questions? >> SPEAKER: Maybe also, like, as a follow-up to this and to understanding more, is that even though if we lay our hands back to this type of technology, there might be a push to decentralize it more, but we’re dealing with money in a lot of cases, and the main thing we’re doing with blockchain and any other type of ledger technologies, we have incentives, and people tend to throw that around Like, the design of incent incentives in many cases is not in terms of a complex system design how we would usually design infrastructure, so my question is what do you think are the roles of the next generation of incentives? Are they still going to be does this work? I think so, let’s see how long this is going to last, because a lot of blockchain projects are pretty much in their infancy, so proof of stake is still trying to be a thing, and we’ve had R chain, which recently is still not pushing, so in a lot of cases, we still don’t know how to do proper incentives. So, that’s my question, what do you think is going to be the next way of implementing incentives? >> SPEAKER: That’s why we’ve started talking to a lot of economists. I think that we need to start to bring some of their rigor into the space Economists do understand how to model and think about incentives in pretty robust, rigorous ways Most of them have been very turned off by this space, because it, you know, sort of some of the ideology and some of the statements and things like that, but I think they’re starting to look at it again. I just attended an event at the University of Chicago that was put on by their economics department, and they were doing excellent work. Most of them are still looking at proof of work in mining, they haven’t moved on to proof of stake or what I think is going to be incredibly important, which are layer two incentives, so we’re kind of trying to have that dialogue with them so that they’re aware of these problems I think one thing is bringing economists and complex system designers, like you said, into the space and explaining, you know, what the open problems are and trying to get them to think about it. Um, this is one area where I’m actually glad that there are, you know, a thousand different coins out there and a thousand different systems, because they can all experiment, and they can go try all these crazy ideas, and we can see, um, hopefully get some good data on what works and what doesn’t work >> SPEAKER: Um, so, that’s a very good, I am even questioning the necessity today of the proof of work model for achieving consensus So, um, we might have a multi-layer consensus model, so I like the quote that they have a set of nodes that are designated as endorsers These endorsers are deciding the read/write set for the rest of the nodes to comply with, and once they all agree, a small group agreesgrees, then the rest, you know, so the question is do you need, um, mining nodes to be driven by a monetary consensus model, or is there a different model, where basically, everybody agrees Earlier this morning, I talked about, well, you know, areblock systems autonomous systems? Because in autonomous system, they have a unified goal You have a completely different model that provides stability, because, again, if your asset is parked in blockchain A and the algorithm is susceptible to manipulation, and somebody just launches a crypto key on the blockchain, and you want to move your asset, and you can’t, you’re not going to be a happy guy, right? You saw my slide on the data gram paper, 1974, I think we are in 1974 of the blockchain. We have a long way to go. If somebody says to you this is going to be done in five years, okay, yeah, let’s talk again in five years We’ll have another symposium in five years >> SPEAKER: Definitely. Any other questions?

>> SPEAKER: I thought it was an interesting statement, that you don’t have a goal center you can call call center you can call when you have a complaint, so my question would be on ownership and, um, if there’s no owner, who holds responsibility? >> SPEAKER: So, here’s — >> SPEAKER: Sorry, maybe to phrase the question a little bit differently, do you believe in fully decentralized, so that there’s no supervising body, that you have a fully autonomous system? >> SPEAKER: So, what if I owned a rack in my basement, and you know who I was, you know what my public key is, so on and so on, and I want to join your community to do mining I don’t mind not being anonymous, and, you know, could that model be used? So, this is the question of, um, in a community that’s processing transactions, do you even need an incentive model? Is there a different model whereby a group of us, you know, all of us spread around the world, a thousand of us get together as a collective, as a consortium, where we donate CPU? And it’s not exactly a mining pool, where our job is to get a thousand copies, a thousand ledgers synced up in as few milliseconds as possible. That’s the service, and if that happens, the group gets paid as a fraction of the transaction Would that work, right? It has to be a thousand physically distributed nodes, not where all thousand is sitting on AWS, it’s ridiculous >> SPEAKER: But, yeah, maybe, from my point of view, I think Eduardo said this morning, um, that access is more important, so Uber doesn’t own cars cars, Air B & B, no real estate, Facebook, no content, so I imagine, could you have an Uber which doesn’t have an office, which doesn’t have, like, you make the system, you make the app, it is as is, and how do you handle maintenance? How do you handle, yeah, how do you handle your complaints? Who do you complain to? >> SPEAKER: Right. Just one second. Sorry, before you try to answer, like, one minor interjection >> SPEAKER: Did I challenge your fundamental beliefs? >> SPEAKER: No, but it exists already Don’t forget about open source software and community-driven software development, because there’s no office you can call , but, like, if you found a bug in the software that’s running on critical hardware, I need to fix it fast, and it actually benefits you. So, there are incentive models for that, and it’s absolutely open source with no, okay, sometimes, an office, but it depends So, sorry >> SPEAKER: Deployment is a different question, right? >> SPEAKER: So, I think sometimes, we forget, decentralization is not an end goal in and of itself, it’s a means by which we are trying to obtain something else I think in bit coin in particular, decentralization is a means by which to obtain money that can’t be changed by anybody else, censorship resistance payments In file systems, it might be a way of obtaining censorship resistance storage. Notice the word censorship resistance seems to come up a lot. It really seems like that’s a core part of the value of decentralization. We actually wrote a paper a couple of years ago investigating a ton of different decentralized systems, IPFS,solid, block stack, steam, even some from the 80s, like the deaspru, the Facebook clone, sort of looking at these systems and what they were trying to do and kind of evaluating, you know, what were the challenges involved there, why did they succeed or not succeed. When it comes to decentralized systems, they are almost always, they perform much worse than centralized systems, centralized systems are always faster, they’re easier to maintain, they’re easier to upgrade, they’re easier to get started with. The UI on centralized systems is usually a lot better, the user experience is a lot better With decentralized systems, it can be very difficult to kind of coordinate all of these things you need to coordinate in order to make this stuff a lot better, and you’re usually operating in a very adsarial environment, so users have to have public keys and encrypt things and check things and verify things Um, that said, the problem that oftentimes happens with centralized systems is sort of what we’re seeing happen now with Facebook and Google, is that they get very big, and they start to say no. They say, no, you can’t build on my API, ninety-three no, I can’t release this information to you,

no, you can’t run whatever software you want, you’re going to run the software I want you to run. Decentralized systems don’t have these problems, because there’s no one to say no. So, that, I think, is one of the core benefits and the reason we’re still pursuing this avenue, is because it’s open, because it lets anybody build on it >> SPEAKER: The currency, which, is, like, going against all of this and still getting the bandwagon of adoption, how do you put, like, all of this together? Because they have ownership, they have a line to call, probably, and it’s growing in market camp, and transactions between borrowers is happening, which is supposed to happen with BTC. How do you see that? What is happening here? >> SPEAKER: Well, the SEC has not sort of given any indication yet as to whether ripple is a security So, um, it’s definitely possible that they might find that it is one, that ripple, the company, is the entity that we’re depending upon for profits with ripple, and so that’s, if that happens, I think that’ll change a lot of the way that ripple sort of operates and runs. Assuming that that doesn’t happen, um, you know, building software for banks is a really valuable thing to do, and they should do that I don’t understand why you need a token in order to make that happen, I think the ripple token is sort of disjoint from the vision that they talk about, um, but, you know, time will tell, we’ll see what they build, we’ll see what happens >> SPEAKER: Okay, so, I think we should start, like, closing the event Somebody doesn’t agree with this (Laughing.) >> SPEAKER: It’s a robot >> SPEAKER: Maybe we have time for one more question. Final question Or two, if they are very quick >> SPEAKER: Hello I’m just wondering, um, you said censorship resistance, is that a good thing? (Laughing.) >> SPEAKER: That is a very complicated thing (Laughing.) >> SPEAKER: That is a very, very complicated thing, and I don’t think I’m prepared at this point to say whether it’s a good or a bad thing I just, I will say that I do think that it depends on what layer of technology you’re talking about That’s it >> SPEAKER: So, I have to say that, first, let’s thank, like, the speakers here that gave us a lot of knowledge (Applause ) >> SPEAKER: And before we all go, I think it’s the right time to make a group picture You know, you are champions to remain here since 9:00 a.m., so let’s make a group picture of the first symposium for this, and I hope to see you next year You know, let’s make it again, right? Let’s get together So, yeah, please, come here, come up

You Want To Have Your Favorite Car?

We have a big list of modern & classic cars in both used and new categories.